Llama Impact Hackathon
Contents
Contents
In November 2024, I took part in the Meta Llama Impact Hackathon in London. Developers travelled from as far as Switzerland and the United States to build under a tight deadline, with substantial prizes on offer.
Our Ambitious Project
My teammates Victor and Chris and I tackled a problem in healthcare. We set out to build an Electron application that could listen to doctor–patient conversations in real time, transcribe them, and generate structured medical notes.
We also wanted the application to suggest questions as the conversation developed, helping doctors avoid missing important points. The same approach could be useful in other professional conversations.
Technical Challenges and Lessons Learned
Our stack combined an Electron frontend with a Python backend, using Llama 3.2 for text processing and OpenAI’s Whisper for speech recognition. The main challenge was not implementing each component, but integrating them into a reliable real-time system.
One of our biggest learning moments came from our approach to integration. We spent too much time developing the frontend and backend independently, leaving the crucial integration phase until around midnight. In retrospect, this was our first major lesson: start with integration. It’s better to have a fully connected system early, even if it’s basic, than to have two sophisticated but separate pieces that you’re trying to join at the last minute.
The hardest technical problem emerged around 4 a.m., when we were decoding live audio while running inference across several models. We could not make the complete pipeline both fast and accurate within the remaining time, so we did not deliver a fully working prototype.
The All-Nighter Debate
The all-nighter was not worth it. We worked until 9 a.m., but code written after 2 a.m. rarely justified the loss of clear thinking the next day. For future hackathons, I would set a hard stopping time and preserve enough sleep to make good decisions.
Reflections and Looking Forward
We did not win, but the experience was valuable. I especially appreciated Victor’s and Chris’s technical skill and determination.
I would take part in similar competitions again. Hackathons expose integration mistakes quickly and force a team to learn under pressure.
Key Takeaways
- Integrate early rather than treating integration as an afterthought.
- Set a realistic scope for the available time.
- All-nighters rarely improve the outcome.
- The problem-solving process can matter more than winning.
- Good teammates make difficult projects enjoyable.
The most useful lesson was operational: build a thin, integrated system first, then improve each component. I have carried that principle into later projects.
If you are considering a hackathon, take part—but get some sleep.