Inspiration
My background is in data science and NLP (natural language processing) in particular, and this problem statement called out to me because it posed a very clear, solvable problem. NLP tools can be used to filter both industry solutions and military capability gaps, and then all that's needed is a good matching algorithm.
What it does
This is currently a search engine which allows the user to type in an industry solution and receive a ranked list of military capability gaps. The capability gaps are sourced from SAM.gov and AFWERX
How we built it
I build this in Python as a Flask webapp using ChatGPT and Codex. I first used ChatGPT to write a pipeline that extracts SAM.gov data using an API key, and once that was built moved to Codex to write the second pipeline for AFWERX data. I then had Codex build a local Flask webapp, confirmed it was working, and deployed to my personal application server.
Challenges we ran into
The main challenge was finding industry solutions. I view this as a second problem which may be solved in a similar way, given enough data sources. I am also using basic NLP to filter for military RFI's right now, and it would be better to use an LLM or better NLP tools like vectorization to improve the search function, but I did not have time.
Accomplishments that we're proud of
I'm proud that this was able to be completed in essentially one day, and we got to an MVP. The app is working, and can certainly be improved upon later.
What we learned
I had not used Codex before. ChatGPT lagged out after using it for a very long chat, and Codex came to the rescue. It is very easy to get started with, and helped enormously in allowing me to complete the app once a basic pipeline was in place that it could learn from.
What's next for Matching Industry Solutions to Military Capability Gaps
The sky is the limit. It would be good to connect with SOCOM and see if they'd be interested in having me improve the tool. Obviously more data sources could be added on the "capability gap" side, and we would need to build out the understanding of the solution space. Another person was working on Problem 5 as well (Dash) and our projects didn't overlap enough to form a team, but we certainly had a lot to talk about, especially with respect to using USPTO data. The app is on a subdomain right now, and it would be good to move that to its own domain, improve the tool, and perhaps generate some traffic. I am considering open-sourcing the project but right now it's better that it's private while I talk to people in industry and see how they'd like to proceed.
Log in or sign up for Devpost to join the conversation.