Latest / Stage Talks with Bellingcat / "Building Tools for Research" with Galen Reich
Transcript
- 0:03You're listening to a stage talk titled Building
- 0:06Tools for Research. This week, we were joined
- 0:08by my colleague, Galen Riesch. Galen is behind
- 0:12many of the Bellingcat built open source tools
- 0:14available today and came in to share his advice
- 0:17on how to get started in tool development, whether
- 0:20you're a coder or someone passionate about accessibility.
- 0:24You can find links to all of the resources mentioned
- 0:27in the talk in the podcast description. This
- 0:30talk was hosted by me, Charlotte Mar, on Thursday
- 0:32the 28th of August 2025 in the Bell and Cat Discord
- 0:36server. So welcome back to our Stage Talk series.
- 0:47This week I'm joined by my brilliant colleague
- 0:50Galen Riesch. Galen's name pops up in many conversations
- 0:55here in the server and within Bellingcat. I was
- 0:58actually surprised to hear that this is his first
- 1:00stage talk. Here's our tech community lead and
- 1:03one of the many amazing people we have behind
- 1:05the scenes creating useful tools and visualizations
- 1:08for our investigations. Here's the mind behind
- 1:11tools that you might be using today, including
- 1:13tools that help you find things like the Shadow
- 1:16Finder tool, a tool that assists with locating
- 1:18a place based on a shadow. and collaboration
- 1:21tools that make working together easier like
- 1:24the search grid generator and the color highlighter.
- 1:27You also may have seen his face in our YouTube
- 1:29tech series, walking you through how to use the
- 1:32command line, or have come across his guidance
- 1:34in our tech community spaces within this very
- 1:37server. Regardless of where you might have encountered
- 1:40Galen, today he is here to discuss the basics
- 1:43of his role, highlighting why you might want
- 1:45to get involved in tool building and what goes
- 1:47into a useful tool for research. We'll explore
- 1:50some of Bellingcat's tools and highlight others,
- 1:53explain what ethical and security considerations
- 1:55you need to make when building a resource, and
- 1:58give you direction on where and how you can get
- 2:01involved. As we talk, please make sure to add
- 2:04your questions in the chat box via the message
- 2:06bubble icon in the top right corner of your screen,
- 2:09and please note within your question if you do
- 2:12not want me to read your username out. Again,
- 2:15this is being audio recorded. So it's really
- 2:17important that you do that if you're not comfortable
- 2:19with me reading your Discord username. All right,
- 2:23Galen, go ahead. Tell us a little bit about tool
- 2:26building. Yeah, thank you so much, Charlie. It's
- 2:30a very generous introduction for the stage talk
- 2:34on building tools for research. I'm going to
- 2:37take a moment just to introduce myself a little
- 2:39bit more and to give you a bit of a sense of
- 2:43my background and where I'm coming. where I'm
- 2:46coming at all of this tool development from.
- 2:49I'm an investigative technologist at Bellingcat.
- 2:53I spent a lot of time working with you wonderful
- 2:57folks in the community. My background is in engineering
- 3:02and research. Before my time at Bellingcat, I
- 3:09was at the BBC where I worked in their research
- 3:11and development department. And before that,
- 3:14I was at the University of Birmingham where I
- 3:17was completing a PhD in radar, so in electrical
- 3:20engineering, which is a far cry from most of
- 3:23the work that I do now. So I'm definitely not
- 3:26a developer by training, although I do now develop
- 3:31a lot of tools and work with folks to develop
- 3:34new investigative approaches. But really, I think
- 3:40For me, the common thread is development of ideas
- 3:45and finding the ideas from the kind of rich pool
- 3:48of thoughts and things that people tell you that
- 3:52could bubble up into a really valuable tool.
- 3:56So that's a little bit about me. What I want
- 3:59to talk about today then is like, why build a
- 4:03research tool? What makes a tool accessible?
- 4:06And finally, what makes a tool usable? We'll
- 4:09talk a little bit about the difference in how
- 4:12I'm using the words accessible and usable there
- 4:15a little bit later on. But yeah, why build a
- 4:18tool? What makes it accessible and what makes
- 4:21it usable? The absolutely most important thing
- 4:26that I cannot overstate, tools must be free in
- 4:32the open source research space. We did some research
- 4:36at Bellingcat a few years ago, some of my colleagues
- 4:39did, and one of the huge takeaways was that any
- 4:43tool that requires some sort of paid access is
- 4:46just like a no -go. Don't even think about it
- 4:49because many people don't have the resources
- 4:51to be able to spend on getting access to a tool
- 4:55on the off chance that it solves their problem.
- 4:58So tools must be free. And I think that this
- 5:01is kind of a philosophy that fits really nicely
- 5:04in with the open source research philosophy in
- 5:08that with a Bellingcat article, our researchers
- 5:11will try and take you through the investigation
- 5:13and show you how the investigation has been done
- 5:16so that you too can replicate it if you want.
- 5:19And that's a little bit more difficult if they've
- 5:22used the super expensive tool. It doesn't mean
- 5:25that we don't use paid tools from time to time.
- 5:28But when thinking about developing a new tool,
- 5:31we have to think that they've got to be free.
- 5:33Now, I have these three key strands that I find
- 5:38useful when thinking about a tool. That's the
- 5:41concept, the utility, and the execution. Is there
- 5:46a good concept? Can it meaningfully advance an
- 5:51investigation? Because if it can't, then the
- 5:53concept isn't very good. If I came up with a
- 5:56tool that increases all the text in a PDF by
- 6:015 % and turns every copy of the word into Comic
- 6:06Sans, that would be a terrible tool. The concept
- 6:09is just bad. It doesn't matter how well the zero
- 6:12utility, and it doesn't matter how well I execute
- 6:15that tool, the concept just isn't there. Moving
- 6:19on to utility, does the tool have utility for
- 6:22people who might actually use it? Will they actually
- 6:27try to use the tool? Does it seem appealing?
- 6:29Does it seem to solve a need? And then execution,
- 6:34is the tool developed well? When people try to
- 6:37use it, like of those people who think it has
- 6:40the utility for them, will they actually be able
- 6:43to use it and actually get the results they need?
- 6:48So these are the three strands that I think I'll
- 6:50unpack those in a bit more detail. So firstly,
- 6:54like, the concept, why build a research tool?
- 6:57So in my experience, there's a couple of strands
- 7:00of like places that tool ideas originate. One
- 7:05is investigative needs. So this is where somebody
- 7:10who is working on an investigation says, hey,
- 7:13there's a problem here. And I think there's a
- 7:16opportunity for solving it through code development
- 7:19or through some other source of tool. Generally
- 7:23speaking, not always, but generally, these are
- 7:25professionally motivated. It's professionals
- 7:27working in investigative environments where they're
- 7:30encountering similar problems day in, day out.
- 7:34And an example of this is with a recent article
- 7:38that went out, not that recent now, but on the
- 7:44One X Bet betting website. Please take a look
- 7:48at that article. I think it's a fantastic investigation.
- 7:50But as part of that, we needed to build a web
- 7:53scraper to go in and pull out a load of videos
- 7:56that were being streamed on this website. And
- 8:00so there, there's a technical need, and we needed
- 8:04to build some sort of research tool that would
- 8:06fill that, that would address that need. And
- 8:10so I went away, worked with a colleague, and
- 8:12we built a scraping pipeline that could run away
- 8:15and grab all of the videos this site streamed
- 8:18over the course of the day. So that's motivated
- 8:21by investigative need and those videos were then,
- 8:24you know, used in the article and used in the
- 8:26subsequent investigation. The other angle is
- 8:30interesting ideas. This is a much more accessible
- 8:34route into building a research tool because it's
- 8:38like conceptually intriguing. I'm sure many,
- 8:40many folks listening will have ideas or will
- 8:44have encountered things and thought, huh, that's
- 8:46curious. I wonder if I could do something with
- 8:49that. The downside is like, will anyone care?
- 8:55An example of this that is very close to my heart
- 8:58is the shadow finder tool that Charlie mentioned
- 9:01in the introduction. So this is a tool that lets
- 9:04you take a photograph of a scene and by measuring
- 9:09the height of an object in the scene and the
- 9:11length of its shadow, if you know things like
- 9:15the time of day and the date, it's able to give
- 9:19you a restricted set of locations where that
- 9:21photo could have been taken. And if you have
- 9:26multiple photos from the same location, you're
- 9:30able to really narrow down the location quite
- 9:32precisely of where that photo could have been
- 9:36taken. And this was a tool that came out of the
- 9:39Discord community. And when I built it, I thought,
- 9:44this is a really cool idea. unites a bunch of
- 9:47really cool things that matter to me, like the
- 9:50kind of physics of the sun and all of the kind
- 9:54of geometry that I spent some time doing in my
- 9:59kind of educational training. It really appealed
- 10:01to me as an idea, but I definitely had that concern,
- 10:04like, will anyone care? And crucially, like,
- 10:07will that tool ever be useful for anything? It
- 10:11turns out it was. After a couple of months after
- 10:14putting it together and making it available to
- 10:17folks came the one expect story. And we actually
- 10:19used it to geolocate a really difficult, difficult
- 10:24one of these live streams, we were able to narrow
- 10:27it down to a city so that folks in the discord
- 10:30could have a look and more have a much more focused
- 10:34search on that area. So these are really the
- 10:39two driving factors. And oftentimes, people will
- 10:45approach with interesting ideas. I think you
- 10:46really have to develop an understanding over
- 10:49time about which ideas will actually be useful
- 10:53in some future investigation. So you've got to
- 10:57have a good concept if you're planning on building
- 10:59a research tool. Right, utility or what makes
- 11:04a tool accessible. So when I use accessibility
- 11:10here, I'm not referring to accessibility in the
- 11:15sense of a website being colorblind friendly
- 11:20or available for use with a screen reader. But
- 11:25rather, I mean the how much prerequisite rather
- 11:31how much prerequisite knowledge is required to
- 11:35access the tool. So I think of this as a ladder
- 11:38of development. So at the bottom of this ladder,
- 11:42you have the kind of messy scripts that you might
- 11:47write as part of an investigation or to test
- 11:50an idea out. This is where tools really start
- 11:52life typically, and they are the least accessible
- 11:55and the lowest effort end of the ladder. So to
- 12:00use a tool at this stage of development, a user
- 12:02has to have Python installed on their machine.
- 12:04If it's a Python script, which for me it often
- 12:06is, they have to have some technical expertise
- 12:09and the confidence to edit the code that you've
- 12:12written for your own specific example. It's a
- 12:16great way to prototype a tool, but it will only
- 12:20ever find a narrow technical audience if you
- 12:22try and publish that as a tool. The next step
- 12:27up, so adding a little bit more effort and making
- 12:30the tool a little bit more accessible, is what's
- 12:33called a library API. This is where you take
- 12:37that script and you give it a small like incremental
- 12:41improvement. So it's still only going to really
- 12:45be useful for people who are technically adept
- 12:49and familiar with the language we tend to write
- 12:52in Python. So I'll refer to tool being written
- 12:54in Python primarily. It will still only be useful
- 12:59for folks who are technically proficient and
- 13:02confident. However, what the library API kind
- 13:06of lets you do is hide away all of the logic
- 13:09of your messy script so that if someone wants
- 13:13to tweak what you've written to suit their investigative
- 13:18need, you've already done the hard work for them.
- 13:21They don't need to edit your code. They can just
- 13:23use your code. So this is a really nice kind
- 13:25of step up. The next step on the ladder is a
- 13:29command line interface. So it's a significant
- 13:32step up in terms of accessibility, but Research
- 13:36from the study that my colleagues conducted a
- 13:38few years ago shows that over 50 % of journalists
- 13:42and open source researchers are not comfortable
- 13:44with the command line. So even building a tool
- 13:49out to a level where it's usable on the command
- 13:52line, and for a bit of context, the command line
- 13:54is the part of the computer that you might see
- 13:57in movies where you see a hacker pulling up a
- 14:00terminal and typing in code. That's the command
- 14:02line. It looks really scary. In reality, it's
- 14:05just a text -based interface to the computer.
- 14:09So a command line interface gives your tool the
- 14:12ability to be used by anyone who has a little
- 14:16bit of confidence on the command line. They don't
- 14:19have to go through the rigmarole of figuring
- 14:21out your API, figuring out your code. They can
- 14:24just use the tool that you've provided. But like
- 14:28I say, many folks are still not comfortable with
- 14:31that. Coming towards the top of the ladder now,
- 14:34we have these kind of couple of user interfaces.
- 14:38So the first that we like to use are Jupyter
- 14:40Notebooks and Colab Notebooks. This is a really
- 14:44nice way to wrap up Python code in a user interface
- 14:50that is maybe slightly more accessible for folks
- 14:52with no programming experience. It still takes
- 14:59a little bit of understanding, but in my experience
- 15:02working with people, if you send them a link
- 15:03to a notebook or either a Jupyter notebook or
- 15:08a Colab notebook, after a little bit of explanation,
- 15:11they're very comfortable using that tool there
- 15:14on out. And there's a bunch of techniques for
- 15:17simplifying the interfaces on a Colab notebook
- 15:20that I recommend kind of investigating if you're
- 15:24interested in developing tools, because for me,
- 15:27This is really the level at which a tool becomes
- 15:30practically useful for people outside of a kind
- 15:33of technical bubble. So I really highly rate
- 15:37Jupyter Notebooks and I really highly rate Google
- 15:39CoLab Notebooks as well. And finally, the kind
- 15:45of the last rung on the ladder, but in reality,
- 15:47it's quite a significant step up in terms of
- 15:50effort. is a full user interface, a full graphical
- 15:54user interface. This is one of the best ways
- 15:57to make a tool accessible to a large number of
- 16:00people, but it can really be a lot of extra development
- 16:02work. And if you're thinking about building a
- 16:07tool with a full interface, ask yourself, is
- 16:10it worth the effort? Is the target audience for
- 16:14my tool? going to need the added convenience
- 16:18of a user interface, or would they be happy with
- 16:20a notebook? If you answer that they would be
- 16:23happy with a notebook, that's what you should
- 16:25do. Don't go the extra mile. Just deliver that
- 16:29tool that meets the needs of your users. And
- 16:34this is something that we try to think about.
- 16:37So, Bellingcat, we publish several command line
- 16:40interface tools. We also... publish a bunch of
- 16:44research notebooks. So we have a repository on
- 16:46our GitHub which contains a bunch of research
- 16:49notebooks that let you perform a variety of tasks
- 16:52that we don't think merit full development of
- 16:57standalone tools, standalone user interfaces.
- 17:04So that's the ladder from a messy editable script
- 17:08at the bottom through a library API through a
- 17:12command line interface, through Jupyter or Colab
- 17:15Notebooks, and finally, a full user interface.
- 17:18The further you go up, the more effort it is,
- 17:21but the far more accessible for the end user.
- 17:26At Bellingcat, we don't particularly advocate
- 17:29a front -end framework if you're thinking about
- 17:31building a user interface. We think that Streamlip
- 17:35seems like a pretty good option. And personally,
- 17:38I have done quite a lot of development in Vue,
- 17:42which is a JavaScript -based front -end framework
- 17:45that can be really good for building user interfaces.
- 17:52Another important thing to consider when you're
- 17:54producing an interface is how are you going to
- 17:57make it secure and how are you going to host
- 17:59it, the whole tool? Because... If you've produced
- 18:04a tool, and I've told you earlier on it's important
- 18:06for it to be free, are you going to have to spend
- 18:11a load of money to host that tool so that other
- 18:14people can use it for free? And the answer really
- 18:16should be no. Building a user interface and hosting
- 18:21a user interface can be expensive, but it doesn't
- 18:24have to be. Streamlit gives you access to a certain
- 18:30amount of compute per month, which means your
- 18:32users don't have to pay to use your tool. It's
- 18:37really powerful. Same with Google Colab notebooks.
- 18:40They don't have any cost to hosting. And if you
- 18:44build a website, I would encourage you to investigate
- 18:49what it means to build a website that is static.
- 18:51So it's one where you just put the files on the
- 18:55internet and all of the complicated stuff happens
- 18:57in the user's browser. It's a really powerful
- 19:00technique for letting users do really complicated
- 19:03things. And it builds security into the process
- 19:08because you're not asking people to upload sensitive
- 19:11files to your server. You just ask people to
- 19:14load it into their own browser. And that's where
- 19:17all of the magic happens. So it's far more secure.
- 19:19And ultimately, it's generally free to host the
- 19:23open source projects using tools like GitHub,
- 19:28which if you're developing open source research
- 19:30tools, and you're not familiar with already,
- 19:32I suggest you investigate GitHub and get more
- 19:35familiar with it. So the final section is execution.
- 19:42So this doesn't refer to how accessible a tool
- 19:47is going to be and how attractive a tool is going
- 19:50to be to a potential user, but rather this is
- 19:54once somebody's opened your tool up, once they
- 19:57want to they've been convinced that this tool
- 20:00could be useful for them, what makes it actually
- 20:02usable? And really, with the exception of it
- 20:07being free, and by free I really do mean free,
- 20:10I can't hammer this home. Freemium isn't sufficient,
- 20:13not a free trial, it really has to be free to
- 20:17get people's buy -in. But the three core principles
- 20:22are it should be simple, so the tool should do
- 20:24one thing really well. Of course, there are many
- 20:28amazing tools out in the world that are very
- 20:30complicated and do lots of complicated things.
- 20:36But my experience in the open source research
- 20:40space is that if you're developing a tool, make
- 20:43it simple. Do that one thing really, really well.
- 20:47Users really relate to tools that meet their
- 20:51need. And you're not trying to sell anything
- 20:54to people. You're not trying to get an annual
- 20:58subscription. You don't have to promise the world.
- 20:59You can say, if you want a tool that puts a grid
- 21:02on a map, I've got a tool for you. If you want
- 21:05a tool that lets you search local council meetings,
- 21:09I've got a tool for you. If that's not what you
- 21:11need, these aren't the tools for you. And a shout
- 21:15out, I think, is appropriate at this point for
- 21:17the Bellingcat Toolkit. a fantastic resource
- 21:22for finding new tools that might meet your investigative
- 21:25needs. So simple, does one thing well. The interface
- 21:30should be clear. You've got to put the user at
- 21:32the heart of the interface. And imagine that
- 21:35when you open up the tool, you have no idea what
- 21:39it's meant to do. And all of the information
- 21:41should be contained within the user interface
- 21:43and within the design of the tool. Coming from
- 21:47an engineering background, This isn't the most
- 21:49natural space for me. But I know that I previously
- 21:54had lots of fantastic colleagues at the BBC who
- 21:57worked on user experience and design. And I feel
- 22:02I learned a lot from them and now I'm a strong
- 22:05advocate for really considering the user's experience
- 22:09and designing accordingly. And finally, if possible,
- 22:15the tool should be quick. This might mean you
- 22:18have to spend a little bit more time optimizing
- 22:21your code or designing your processing pipeline
- 22:26differently. But ideally, a user should be able
- 22:29to press the button and get results nearly instantly.
- 22:33Because otherwise, if you press a button and
- 22:36see a loading bar or a spinning wheel of doom,
- 22:42it really erodes trust. folks will start to think,
- 22:46oh, have I done this right? Have I made a mistake?
- 22:49And in the world of these open source research
- 22:52tools, if a user makes a mistake that causes
- 22:55your tool to fail, you don't want that pathway
- 22:59to be very similar to using your tool normally.
- 23:02You want them to get a really clear error message
- 23:05and a nice instruction to please contact you
- 23:07as the developer and let you know what they were
- 23:09trying to do when it went wrong. really inviting
- 23:12feedback and inviting development is such an
- 23:15important part of making an open source research
- 23:18tool usable. So yeah, to recap, I think you need
- 23:28that great concept. You need the idea to be a
- 23:30really solid one. Put yourself in the shoes of
- 23:33an investigator if you don't work in that space
- 23:35and think, is what I'm offering valuable to them?
- 23:39Does it solve a problem that they either have
- 23:42or could conceivably have in the course of an
- 23:44investigation? Then think of how much effort
- 23:48is reasonable to put in to make this tool have
- 23:52utility because you don't want to be spending
- 23:56months and months and months developing a tool
- 23:58that nobody uses. Build something, put it into
- 24:01the world and see what people think. And finally,
- 24:05when you're building it, when you're executing
- 24:06that tool, that tool development, make sure that
- 24:11you put the user at the heart and really design
- 24:14the experience so that the entire tool comes
- 24:19together for your user. And with all of those
- 24:21things, you'll build a fantastic research tool
- 24:23and people will love using it. Charlie, I think
- 24:27that leaves me to hand back to you. Yes, thank
- 24:30you so much, Galen, for that really clear and
- 24:36knowledgeable. presentation. It's really amazing
- 24:40to see how you break down the parts of tool development.
- 24:45It's been really helpful actually for a number
- 24:47of people in the chat. There are some common
- 24:50themes emerging in the chat though, so I wanted
- 24:52to ask you first about those. Do you have to
- 24:56have coding know -how to get involved with tool
- 25:00building? And if so, in terms of maybe leading
- 25:04tool development? What other ways can you get
- 25:07involved with tool development that might not
- 25:09be so focused on the coding section of development?
- 25:15Absolutely. So the short answer is no, you definitely
- 25:18don't have to be a coder to get involved. I think
- 25:23it definitely helps to have the more familiarity
- 25:31with coding you have. after the easier time you'll
- 25:36have when you're exposed to these quite technical
- 25:39spaces, but it's definitely not a barrier. There
- 25:43are a huge number of ways that folk can get involved.
- 25:45One thing that we always find helpful is when
- 25:51folk who don't have a technical background or
- 25:55technical experience try to use our tools. If
- 26:01you're wanting to get involved with tool development,
- 26:03the easiest way, I would say, is to try out a
- 26:07bunch of tools and see if they work. And if you
- 26:10encounter a problem, reach out to the developer.
- 26:15So we have spaces in Discord here where if you
- 26:22encounter a problem, you can shout about it.
- 26:26And really, like... feel free to test the tools
- 26:28to like their breaking point, because as a developer
- 26:31that works on some of these tools and supports
- 26:34other developers in building these tools, you
- 26:36know, you do your best to think about what people
- 26:38are going to do with the tool. But ultimately,
- 26:42it's not about us developing it. It's about you
- 26:45using it. So you don't have to have any coding
- 26:47knowledge to use many of the tools we produce.
- 26:50But if you try and use it and we failed in the
- 26:52execution step, then we failed and we need to
- 26:55know about that. So a lot of the, for me, the
- 27:01joy, so we just published a tool today, more
- 27:04or less, and I've had some feedback from folks
- 27:07on Blue Sky about tweaks that could make the
- 27:12tool more usable. And what I've done is I've
- 27:14taken those and I've noted those down in the
- 27:16tools repository on GitHub. So if you're comfortable
- 27:21enough to create an account on GitHub, there
- 27:23are ways to get involved with the conversation
- 27:26around tools, even without having to be a developer.
- 27:32And if you head to Bellingcat's GitHub, which
- 27:35is github .com forward slash Bellingcat, there's
- 27:38a big section on the landing page there that
- 27:41explains how you can get involved with Bellingcat's
- 27:44tools specifically. So I hope that answers the
- 27:47question. Within the chat, we've got people saying,
- 27:50for example, Sarah, our moderator. She says,
- 27:53you can be a technical writer as a contributor.
- 27:56People love us. And that is really true. That
- 28:01is being able to really describe how to use a
- 28:05tool is, is part of the reason why our Bellingcat
- 28:08toolkit exists. Because not enough people do
- 28:12this when they build tools is make, you know,
- 28:15really good documentation. So people understand
- 28:18how to one use it and. the benefits and the disadvantages
- 28:22of each tool as well. So that's in the chat as
- 28:25well that I linked the toolkit. I've also just
- 28:28put in the chat some links to spaces within the
- 28:34server. Gelin mentioned the GitHub, but we also
- 28:38have within the server tech contributors, which
- 28:41is a place to find ongoing tool ideas and to
- 28:45pitch those tool ideas and to find people who
- 28:48want to build together. And tools and sites is
- 28:51a place where people share often useful tools
- 28:53that they've come across. And tool support is
- 28:56that place within the server where you can ask
- 28:59for help with a particular tool or flag something
- 29:02that's not quite working, particularly with a
- 29:05Bellingcat one if you're not comfortable or aren't
- 29:08familiar with GitHub and the way that that works.
- 29:12Just building on this slightly. Somebody said
- 29:16that they'd be comfortable picking up an idea
- 29:20from a list of ideas people had, but didn't try.
- 29:23Is there anywhere like that? So we've just mentioned
- 29:25the tech contributors within the server. Is there
- 29:28anywhere online where people do say, oh, we have
- 29:33this idea for a tool, but don't necessarily have
- 29:35the resources to build it? Is there somewhere
- 29:37like that where you can get involved in those
- 29:39kinds of communities? You mentioned already the
- 29:42tech contributors space. There's a channel in
- 29:45there for working on working on new ideas. I
- 29:50think it's called welcome and info, which is
- 29:52perhaps not the best name, but it can be used
- 29:56for developing ideas. And within tech contributors,
- 30:00it's a forum channel. So there are a bunch of
- 30:02different ideas that are in there. Oh, no, there
- 30:06is a I misspeak there is a a section of tech
- 30:10contributors called tool idea workshop. So that's
- 30:12where people might be discussing new ideas. I'll
- 30:15be honest, it's a little bit of a stagnant channel
- 30:17right now. So if you're if you're interested,
- 30:20if you're excited by this, please go dive into
- 30:23tool ideas, workshop, share your ideas and really
- 30:26like use that as a space to collaborate with
- 30:31others. We do have a We do have a space with
- 30:36some long -standing challenges on the Bellingcat
- 30:41GitHub. That too is a little bit stale now. I
- 30:44mean, it's something that I've been working on
- 30:46a little bit in the background. And I hope that
- 30:50at some point soon, I'll be able to release a
- 30:52kind of list of like a shopping list of ideas
- 30:56that we at Bellingcat think are really valuable
- 31:00and would be really cool enhancements to the
- 31:04open source research world, but they just don't
- 31:06exist yet. That's something that I'm working
- 31:09on in the background and hopefully I'll have
- 31:11something to share publicly on that in the not
- 31:13too distant future, but we'll have to wait and
- 31:17see. Yeah, I just also want to thank John in
- 31:21the chat, who's been also putting in a bunch
- 31:24of useful links, which I'll put in the description
- 31:26of the podcast as well for those who do want
- 31:28to learn the command line and coding as well.
- 31:31So thank you so much for sharing those. And it's
- 31:34really valuable, as Galen said, to know coding
- 31:37for these kinds of tasks, but it's not exclusive.
- 31:41Joanna in the chat, a good colleague and the
- 31:44person, the brain behind the toolkit has asked
- 31:48Galen, what has been your favorite tool building
- 31:50experience so far? Why was it so fun? So it's
- 31:54a great question, Hannah. I think there are,
- 32:00I've enjoyed building lots of different tools.
- 32:03Um, for different reasons, I think, um, and I
- 32:07need to not let recency have an impact. I recently
- 32:11launched the council search tool, which is a
- 32:13UK and Ireland focus tool for finding verbatim,
- 32:18um, verbatim quotes from council meetings. Um,
- 32:23I'm very proud of that and it's very, um, it's
- 32:25very recent, but I think my, my favorite tool
- 32:30development experience has been the shadow finder
- 32:32tool. Because it came out of an idea in the community
- 32:37that I hadn't thought of before, I was aware
- 32:41of some of the methods that you can use for geolocation
- 32:45and chronolocation with the sun. And I was aware
- 32:47of websites like Suncalc. But the idea of running
- 32:51Suncalc backwards hadn't ever occurred to me.
- 32:54And so when Gabor Friesen came into the Discord
- 32:58and shared the idea, I thought, oh, that's really
- 33:02smart. That's really smart. It's probably not
- 33:04going to be useful all the time for every project,
- 33:07but maybe one in a thousand projects is going
- 33:13to hit against a brick wall where the geolocation
- 33:16is just super difficult. But you have this other
- 33:20kind of information. And I think that was the
- 33:23most rewarding for me, both because it kind of
- 33:25scratched that kind of intellectual itch. but
- 33:31also because it seemed to be a really valuable
- 33:35idea that came out of the community. So Gabor
- 33:37and I collaborated on building the tool, me doing
- 33:41the coding primarily, but then we would kind
- 33:43of iterate. I would share a version with him
- 33:45and we'd talk about what works and what didn't
- 33:48work. So I think having a partner in development
- 33:53really made that experience stand out for me.
- 33:55I'm just going to take the opportunity as well
- 33:57to shout out about the grid search tool. This
- 33:59is at grid .bellingcat .com. It's such a simple
- 34:02tool. It also came out of conversations with
- 34:05the community, but there was less iteration because
- 34:09it was such a simple need. Plainly, the idea
- 34:12is if you're doing a geolocation with five people
- 34:15trying to work systematically in an area, how
- 34:18are you going to coordinate the workload? What
- 34:21the grid tool lets you do is it lets you download
- 34:24a grid that goes over a map in Google Earth.
- 34:29And so you can say, right, I'm taking this section
- 34:32of grid cells. I'm taking this section of grid
- 34:34cells. And you can divide work and mean that
- 34:37people aren't double checking the same area.
- 34:40And yeah, it can help you keep track of areas
- 34:42you've already checked. So it is useful in like
- 34:47a few fronts. But it's such a simple tool. It
- 34:49just creates a KML file that you can download
- 34:52and open in Google Earth. It's really like the
- 34:55tech equivalent of getting a pencil and a ruler
- 34:59and drawing a grid on your map. Well, we're shouting
- 35:03out tools. You mentioned earlier that a new tool
- 35:06came out today. Do you want to share a little
- 35:08bit about what that tool is before we delve into
- 35:11other audience questions? Yeah, absolutely. You
- 35:15must cut me off if I rabbit on about it. Yeah,
- 35:19this is a tool that I've been working on lately
- 35:21and it's just reached the state where it's ready.
- 35:27It's a council search tool. This was born out
- 35:32of conversations with local journalists in the
- 35:36UK who one of the kind of classic local journalism
- 35:40tasks is attending council meetings or watching
- 35:44in the modern day, watching live streams of council
- 35:47meetings to see if anything scandalous happens
- 35:51or if there are any discussions about new developments
- 35:53or interesting things for a local journalist
- 35:58to cover. And one local journalist I was talking
- 36:03to said, yeah, I spent I spent every evening
- 36:07of my life doing this. And one day I got a real
- 36:09big scoop because they forgot to turn the recording
- 36:12off at the end. And I thought, oh, that's really
- 36:15interesting. And that's great if you've got the
- 36:19resources to do that. But I know that in the
- 36:21UK, at least, local journalism is one of those
- 36:24areas that's been consistently underfunded and
- 36:27undersupported over recent years. So anything
- 36:31for local journalists and local democracy researchers.
- 36:36to help them cover local council meetings seemed
- 36:40like a valuable addition. And what this tool
- 36:42does is it goes away and in the background, it
- 36:45grabs the transcripts associated with those live
- 36:48streams that councils put out there into the
- 36:52world, and it makes those transcripts searchable.
- 36:55So if you go to the search tool now, you would
- 36:59see a big search bar and you can type in things
- 37:02like off the record or That's outrageous. Or
- 37:08you could type in specific company names that
- 37:11you're interested in. And you can then start
- 37:13to see all of the meetings in which those terms
- 37:18have been mentioned. And the search tool gives
- 37:20you a little preview of the section of text.
- 37:23And crucially, there's a button that takes you
- 37:26to the source video. And if you click that button,
- 37:29it's going to take you not just to the page where
- 37:31the video is, but it's going to take you to exactly
- 37:33correct timestamp for the part that you searched.
- 37:39I'm really hopeful that this will become a powerful
- 37:42tool for local democracy researchers and local
- 37:48journalists. It's all open source and at the
- 37:52moment I just have written code to support one
- 37:55council multimedia provider and I have an aspiration
- 38:01that other folks who are interested and want
- 38:03to get involved will go out there and look at
- 38:06their local council or if you're not in the UK
- 38:09or Ireland, you'll look at a council that you've
- 38:12come across as part of an investigation or just
- 38:14the first one that came up on your Google search
- 38:16and you'll see what format they've used for their
- 38:20meeting streams and you'll see if you can reverse
- 38:23engineer that in a way that can add support for
- 38:26that council and to other councils that use the
- 38:28same provider into this tool. A current outstanding
- 38:33task is to support YouTube, which many councils
- 38:37use. I've not done that yet. And I think it would
- 38:41be really exciting to be able to support those
- 38:47many, many councils that use YouTube to archive
- 38:50their council meetings. One of the audience members
- 38:53has just put, reminds me of Filmot, a transcript
- 38:56search for YouTube. So maybe that's something
- 38:58to check out. Sarah says, that's sick. What a
- 39:02time saver. As someone who used to sit in council
- 39:06meetings as part of her job, I 100 % agree. They
- 39:10were hell. John asks, is this on GitHub as well?
- 39:15Because I know that most councils in the Netherlands
- 39:17publish this as well. Yes, it is on GitHub. So
- 39:21it's open sourced. So it, we have a hosted backend
- 39:31service for it. So we host the database for this
- 39:34tool. And that does all of the kind of scraping
- 39:37on the backend. But the front end is open source.
- 39:42The backend is open source. And if you, did you
- 39:49say it was the Netherlands? In the Netherlands,
- 39:53many councils publish. Yeah. I think adding different
- 39:58country support would be something that I would
- 40:01be excited to do. It's just not been something
- 40:05that's been in the initial scope of the project,
- 40:07but there's absolutely no reason from a technical
- 40:09perspective why that wouldn't work. Or indeed,
- 40:12if you wanted to host your own country -specific
- 40:15version of this, that would be amazing too. Please
- 40:20have a little play with that. Let us know if
- 40:22you come across any issues. You know the spaces
- 40:25in GitHub and the spaces are then here to flag
- 40:28that now. It's really, really important. And
- 40:32yeah, we wanted to really announce that today
- 40:36just so that you had a little tool, a new tool
- 40:39to play with after this talk. I'm just taking
- 40:42it back to building your own tools though, away
- 40:45from the current available tools. We had a question
- 40:49earlier about how EU nations are currently moving
- 40:54to free open source software away from Microsoft
- 41:00Office, for example. And they asked about what
- 41:04it meant for cyber security. But I had a wider
- 41:06question on that in terms of actually building
- 41:09open source tools. Are there any security issues
- 41:13that you need to consider when building and opening
- 41:18up your code to the world. This is one of the
- 41:22reasons why I'm quite a strong advocate for using
- 41:25those kind of Colab notebooks or producing a
- 41:29website that is static and runs entirely in a
- 41:31user's browser. Because in doing that, you kind
- 41:35of circumvent many of the security challenges
- 41:39of deploying a service like we've done with the
- 41:42council search tool. because then you have to
- 41:46make sure it's secured and you have to have some
- 41:49understanding of how to do that effectively.
- 41:55So yeah, there are definitely security concerns
- 41:59in publishing a tool, but if you make it so that
- 42:03it only runs on a used device and you're not
- 42:08loading malicious code onto people's computers,
- 42:12that's a really good thing. As long as you're
- 42:15not the security problem is I think what I'm
- 42:18saying. Following those tips circumvent a huge
- 42:22number of those issues. And then on the idea
- 42:28of open source tools were generally so there's
- 42:33a subtle distinction. that we encounter really
- 42:36frequently in the Bellingcat tool space is the
- 42:39difference between open source research and open
- 42:42source technology, open source code. So they're
- 42:46very similar, but slightly different. So with
- 42:50open source research, that means we're using
- 42:53open sources of information to conduct our investigations.
- 42:58And there an open source could require payment.
- 43:02So in Bellingcat investigations, we will sometimes
- 43:05use imagery from the satellite company Planet
- 43:07as part of our investigation. Now, that's open
- 43:13source in the sense that anyone could pay Planet
- 43:16for that imagery. And we also then publish the
- 43:20imagery server, kind of check our work. But it's
- 43:23not open source in the free and open source software
- 43:27sense, where all of the code is public and available
- 43:31for people to iterate on and improve. Now, incidentally,
- 43:34I think both sets of ideas are really compatible.
- 43:37It's why we publish so many of our tools in an
- 43:40open way, because it just fits the, it's a really
- 43:45good intersection between technical folks and
- 43:47research folks that I think can lead to really
- 43:51kind of fruitful relationships. And I think more
- 43:54and more as institutions move towards kind of
- 44:00open source tools, I think or I hope that the
- 44:04community around open source continues to grow
- 44:07and it helps just everyone build better tools
- 44:10in the long run. This is a question that comes
- 44:12up quite a lot when we're talking about tool
- 44:15building. Obviously people tend to think a lot
- 44:18at the minute about AI, one of the most common
- 44:25AI tools. that are kind of emerging at the minute
- 44:30is kind of these verification checkers. I had
- 44:33a conversation with someone very recently who
- 44:36is keen to build a similar one, and they're asking
- 44:39us about the types of disinformation and misinformation
- 44:41that we encounter, because that's something they're
- 44:43looking into. But Aries in the chat has asked
- 44:47an interesting question about how tools specifically
- 44:52are preparing, particularly when it comes to
- 44:56detection or counter misinformation tools, how
- 44:59they're preparing to cope with new AI misinformation
- 45:04and disinformation. What kind of architectural
- 45:09feature level innovations are on your mind to
- 45:11help investigators detect and counter misinformation
- 45:13while still leveraging AI effectively is the
- 45:16question. But I don't know if you've ever come
- 45:18across a tool that is 100 % reliable when it
- 45:23comes to identifying AI, I haven't. I know we've
- 45:29recommended some in the past with additional
- 45:34verification techniques, of course. But yeah,
- 45:37I don't know. Do you want to speak a little bit
- 45:39to that space that currently there's a lot of
- 45:42tool builders, developers kind of looking at
- 45:44and what that particular space looks like? Definitely.
- 45:50And I think there's a few strands to it as well.
- 45:52So I'll try and... speak to each of them as I
- 45:56see them, and hopefully that's useful. So I think,
- 46:01firstly, there's the whole space about like AI
- 46:03detection tool. AI detection tools are going,
- 46:09it's going to be a constant battle between AI
- 46:13detection tools and nefarious AI trainers. Of
- 46:18course, there may be cooperative solutions between
- 46:20like larger platform, but AI detection is like
- 46:25an adversarial exercise, will always be a kind
- 46:29of shifting ground. And I don't think, I don't
- 46:33think there will ever be a tool that delivers
- 46:39good results always and is consistent, like is
- 46:45ahead of the curve for a long period of time.
- 46:47It might be that a tool that is very effective
- 46:49one month. is going to be obsoleted with the
- 46:52next model release and its developers will have
- 46:56to go back to the drawing board to improve their
- 46:58detection model. But I think sometimes I think
- 47:03we risk focusing on detection tools when actually
- 47:10the kind of ecosystem of misinformation is perhaps
- 47:17a much more fruitful place to investigate. So
- 47:19I think to stay ahead of synthetic media and
- 47:22those advanced obfuscation approaches is almost
- 47:27to forget about the technical element. If you
- 47:31see an image online that is surprising or shocking,
- 47:39follow your nose and follow your investigative
- 47:42instincts and see, well, if there is a huge fire
- 47:48in this location, Why are all of the images this
- 47:51exact one angle? Can I find multiple images on
- 47:58social media posted by different sources, like
- 48:01beginning to build up a picture of the misinformation
- 48:06landscape around a particular event, I think
- 48:08is perhaps a more useful approach than putting
- 48:11an image into an AI image detector. Of course,
- 48:17there's a place for those detectors, but I think
- 48:20as the technology moves on and as things become
- 48:25more and more convincing and it becomes increasingly
- 48:27difficult to separate reality from a generation,
- 48:32I think it will become those contextual clues
- 48:37that provide the really kind of richest information.
- 48:41So I apologize, it's kind of a non -technical
- 48:44answer to the question and it's not a space that
- 48:46I'm a particular expert in. I know my colleague
- 48:49Kalina would probably have a much more intelligent
- 48:51answer to how to spot misinformation, particularly
- 48:58AI -generated misinformation. But I think it's
- 49:04an honest one. I don't think all problems have
- 49:07a tech solution. And that's probably quite an
- 49:10important lesson for those of you thinking of
- 49:14going into kind of or being involved with like
- 49:17tech development in this space is sometimes there
- 49:20are very human solutions to these problems as
- 49:23opposed to technical ones. Yeah, I can count.
- 49:25I can't even count on my hand the amount of times
- 49:28I've been asked by a journalist when I've been
- 49:31doing talks on disinformation. Is there like
- 49:34a tool that fixes this? It's like the answer
- 49:36is no. That's not the answer. The tool isn't
- 49:40the answer, as far as I'm aware. So I agree with
- 49:44you there. I wanted to ask though, in terms of
- 49:46when we're talking about updating tools and making
- 49:48sure that they're advancing enough with the tech.
- 49:53We've had some tools in the past, spelling out
- 49:55tools break when social platform policies have
- 49:59changed, right? How do you keep on top of that,
- 50:02particularly when you're building search tools,
- 50:04search across specific platforms? a tremendous
- 50:08amount of effort, frankly. I think there's not
- 50:13really a good way around it. If a tool that you're
- 50:19building relies on somebody else's service, then
- 50:25that is, as part of developing that tool, you
- 50:28have to understand that if you want to keep that
- 50:31tool working in the long run, you have to...
- 50:35you have to accept the possibility that that
- 50:37person could change how they're operating. This
- 50:39is particularly true in the world of scraping.
- 50:41So if anyone has tried to do any social media
- 50:44scraping in the last five years, frankly, you'll
- 50:49have found a plethora of tools, some of which
- 50:53are very good one week and terrible the next.
- 50:56some of which are kind of mediocre all the time,
- 51:00and some of which are just kind of terrible.
- 51:01But as far as I'm aware, there's no like consistently
- 51:04fantastic social media scrapers, because those
- 51:07social media platforms are constantly shifting.
- 51:10And so the amount of work required to keep up
- 51:12with them is significant. It doesn't mean it's
- 51:18not worth doing, but it's definitely a consideration
- 51:20when building a tool is thinking like, how, how
- 51:23do I keep this supported in the long run? It's
- 51:26why with a tool like the grid search tool, I'm
- 51:29super happy because that will, I hope, fingers
- 51:31crossed, require zero maintenance along because
- 51:34it does what it says. It does that one thing
- 51:37really well. And it lets you download a file
- 51:39format that's barely industry standard, which
- 51:42is a KML file format. So there's no shifting
- 51:49ground there. When it comes to tools that rely
- 51:51on third party services. That's a feature of
- 51:54the landscape. And I think you either accept
- 51:57that your tool is going to take a lot more of
- 52:00your time or you accept that at some point it's
- 52:02going to break. And that's just how it is. Yeah.
- 52:05And as you said at the beginning, I think simplicity
- 52:08is often key when you're thinking about applications,
- 52:11particularly. Do you often, within tool documentation,
- 52:14do you often have to assess? What devices can
- 52:18use your tools and in what regions tools are
- 52:21applicable as well? Is that something that's
- 52:23really useful to add into documentation, for
- 52:25example? If you have any, the general rule with
- 52:28tool documentation is if you have some insight
- 52:31that you think is useful to someone else, put
- 52:33it in the documentation. On a personal level,
- 52:36I don't think I have a tremendous amount of insight
- 52:38into what is... what are useful to folks in different
- 52:44regions. It's something that I have in mind and
- 52:47would like to better my understanding of so that
- 52:50we can build better tools that are more portable
- 52:52and more useful to a larger number of researchers.
- 52:58In terms of devices, it's a really good question
- 53:01because I think more and more journalists are
- 53:08doing things on their phone. more and more researchers
- 53:10will be doing things on the go. You might be
- 53:13on the train doing a little bit of research.
- 53:16So finding ways to make your tool suitable for
- 53:20mobile is like a really essential thing in modern
- 53:23landscape. And that is a downside of the Colab
- 53:28Notebooks, Jupyter Notebooks that I talked about
- 53:30earlier is that they are not very accessible
- 53:32for mobile device. You immediately cut out a
- 53:36significant fraction of your users or at least
- 53:40your users a significant fraction of the time
- 53:42when they're on the go. It's one of the things
- 53:46that I really like about using Vue, that's V
- 53:49-U -E, the JavaScript framework for front -end
- 53:52development, is kind of implicit in the design
- 53:57is a mobile -first approach, which means that
- 54:01your site will be good on mobile. So the council
- 54:05search tool should work on mobile. And you should
- 54:10have a perfectly good experience there. So thinking
- 54:13of that and testing for if your tool works on
- 54:17mobile is a really good thing to do. And I know
- 54:23it's something that we do. This is maybe a slight
- 54:25aside from tools, but it's something that we
- 54:27do when thinking about article content for Bellingcat
- 54:30stories is does the article look good on a tiny
- 54:35mobile phone? I'm just going to pull up my responsive
- 54:42design mode in my browser to give you a sense.
- 54:45So the size of phone screen that we test for
- 54:48if we're doing some interesting kind of novel
- 54:52experience in an article is 320 by 547 pixels.
- 54:58That's the screen size or the functional screen
- 55:02size of an iPhone 5. which is tiny. I can't overstate
- 55:10how small that is. So when you're building a
- 55:13tool, think about what's it going to look like
- 55:15on a tiny, tiny screen? What's it going to look
- 55:18like if that person on a tiny, tiny screen turns
- 55:21their phone into landscape mode? How is this
- 55:24tool going to look? Is it even going to be usable?
- 55:25Do we just have to tell them to turn their phone
- 55:27around? But if you can do that, that's such a
- 55:31valuable addition to a tool. If you can add those
- 55:37graceful degradations as people's screen sizes
- 55:41get smaller, as it becomes more and more difficult
- 55:44for them to access the functionality of the tool,
- 55:47let it lose its functionality in a graceful way.
- 55:51Tell the user that it would be better if they
- 55:54found a larger screen or turn their phone around.
- 55:57And make sure you do that. early on in the tool
- 56:00interface planning process. Because there's nothing
- 56:03more frustrating than if you've built this beautiful
- 56:06tool interface and then you test it on mobile
- 56:08last minute and it doesn't fit. That is gutting.
- 56:13So that would be my advice is to test it early.
- 56:17I'll let you know what I've cracked that one.
- 56:19Yeah, as you can with the council search tool,
- 56:22please do test that out for Galen on mobile as
- 56:24well. We are coming up to the end of the hour,
- 56:27so I wanted to ask you one last question. You
- 56:29mentioned at the beginning that most of the open
- 56:34source tools are free. If somebody does want
- 56:38to make a business out of this, make money, is
- 56:41that even possible within the open source formats
- 56:46or is that something that is just so anti the
- 56:48ethos that you don't recommend it? I don't recommend
- 56:53it, not because it's anti -the -ethos. So for
- 56:57instance, the Mozilla Foundation, they release
- 57:01the Firefox browser, which is an open source
- 57:05web browser. But they are also a commercial entity.
- 57:10And as I understand, part of their business model
- 57:13is providing support to organizations that want
- 57:16to use the open source tool at scale. And that's
- 57:19quite a common model. see in the open source
- 57:24software space is an organization publishing
- 57:26something for free and then selling services
- 57:29to help businesses use it at scale. So that kind
- 57:36of infrastructure support for larger, more complicated
- 57:39tools is an avenue where you can make a business
- 57:45out of it. I'm the wrong person to ask about
- 57:51developing proprietary tools because it's a different
- 57:58trust architecture to the one that we like using
- 58:01at Bellingcat, which is one where the method
- 58:04is very clear, the tools are very clear, and
- 58:07open source tools make their method implicitly
- 58:10transparent. If somebody wants to sell a tool
- 58:15that does a thing and you're not going to tell
- 58:17me how I do it, it makes it much harder for me
- 58:20to justify using it in an investigation. It's
- 58:22not impossible. Sometimes tools will unlock a
- 58:28level of analysis that is tremendously helpful
- 58:31and unsticks an investigation, but it then becomes
- 58:35a conversation about whether that tool should
- 58:37be used or not. And so I'm the wrong person to
- 58:40advocate for a paid tool ecosystem. Keep it open,
- 58:46keep it free is the message there. Thank you
- 58:49so much, Galen, for your time today. And thank
- 58:52you to everybody who's joined the audience. Please
- 58:54do check out the tech contributor space within
- 58:57the Discord. I'll pop it again in the chat now.
- 59:02It is a space for you to chat with Galen and
- 59:06others from our team and each other about possible
- 59:10tools. And you can see some of the ones in there
- 59:12that Galen is already talking to people about.
- 59:15And do check out our Bellingcat GitHub as well
- 59:18that has tons and tons and tons of useful tools
- 59:23on day. Experiment with them. And if you find
- 59:25an issue, please flag it to us so we can fix
- 59:27it quickly. Thank you again, Galen. This recording
- 59:32will hopefully be up on podcast platforms, which
- 59:36will be posted in here by this weekend. And we'll
- 59:39be back in two weeks time with another stage
- 59:41talk. But for now, thank you and goodbye. Thank
- 59:45you, Charlie, so much for hosting and moderating
- 59:48and keeping track of all questions. As always,
- 59:50you do a fantastic job. Oh, the flattery. The
- 59:53conversation's blowing. Thank you very much.
- 59:58Thank you for listening to the Stage Talk. If
- 1:00:02you'd like to catch a Stage Talk live where you
- 1:00:05can ask the guest questions, join the Bellingcat
- 1:00:07Discord server by visiting www .discord .gg The
- 1:00:15music you've heard is titled Dawn by Newer Self
- 1:00:19and is courtesy of Artlist.