Skip to content

All writing

Going for glory: A hackathon story

4 min read

Intro

4 hours of sleep, 20 hours of laser-focused work, and one brilliant product idea later. This is a story of how my friend and I participated in the Hack the Change 2025 hackathon and almost won. Almost.

I've done two hackathons before, both with teams of five. This time was different; I was confident enough to win, so my friend and I kept it to just the two of us instead of the maximum five. Less distraction, more focus.

The prompt: How can we leverage technology to make cities and human settlements inclusive, safe, resilient, and sustainable?

And with that, the brainstorming began.

The Idea

The prompt asked: How can we leverage technology to make cities and human settlements inclusive, safe, resilient, and sustainable?

Our answer was simple: start with the humans living there.

Physical proximity creates something digital spaces can’t replicate — real connection.

Our final idea was an anonymous chat app, where there would be a physical QR code that you would need to scan to enter the chat room. When you enter the chat room, you will have an anonymous name so you can talk freely. Only those in that area—whether it be a room, a lounge, or an area in a library—could enter, and once you walked away from that area, you would automatically exit the room. This keeps the chat room exclusive to those who are physically there.

The idea behind this app is to reduce the friction in communication with those around you. It could act as an icebreaker and lower tension, helping people eventually interact in real life. This is especially beneficial for introverts who struggle to talk to strangers. The anonymous aspect lowers the barrier to starting conversations.

On top of this foundation, we added features that synergize well with the core concept. The biggest one was AI, with room-specific context.

Imagine a professor teaching a complex topic with little information available online. If a student misses something, they could anonymously ask a question with a room tag. People or AI can answer. AI would retain the context from the most upvoted responses to answer future questions more accurately.

Another scenario: asking the AI where a bathroom or water fountain is if the “room” represents a library floor or lounge. This brings AI into daily life to make navigation easier and fill in missing context so people can stay present.

The scale of this project was ambitious for two people in 24 hours, but we rose to the challenge. Even though we didn’t win, we came very close to achieving our goal.

The Final Product

What We Built

Our final product worked pretty close to what we imagined. We were able to scan the QR code, enter rooms with anonymous names, and use room tags. The AI responded and retained context from threads. We added reCAPTCHA to prevent bots, since an account isn’t required to join a room—but you do need one to upvote comments.

We also added a monitoring function that listens to all messages in every room and deletes inappropriate ones. There’s also Google Auth login.

There are definitely areas for improvement, but in the end, we hosted both the backend and frontend with everything working together. The only major goal we didn’t achieve was automatically removing users when they physically leave the area.

Future Features We Envisioned

There are many ways this app can grow. Some ideas:

  • Games you can play with others in the room

  • Custom tags like emergency alerts, which would broadcast to nearby rooms

  • Limited communication between rooms, while still keeping the focus on local conversation

  • Various improvements to make the product more special

With just two people and limited time, we could only do so much.

The Challenges

Our biggest challenge was staying positive. I tried my best not to get distracted until it was over. I don’t think I’ve ever been this laser-focused for so long.

We worked non-stop, slept for 4 hours, then worked again until the end.

A tough moment was when we started rethinking our idea mid-hackathon. Later, while preparing the submission and presentation points, we kept iterating so much that we started finding flaws everywhere. We over-optimized our messaging to the point of brain fog.

We joked that it was like AI generating a great essay, but if you make it train on itself too many times, it degenerates into nonsense. That was us.

Also, trying to think clearly on 4 hours of sleep did not help. Once we realized what was happening, we stopped and meditated until the interview.

In the end, we didn’t win. I believe our biggest weakness was not marketing the product well in the submission video and not presenting strongly in the final interview. However, I’m still proud of what we accomplished. I genuinely believe in this idea, and whether built by me or someone else, something like this could be big.

I’m just not sure if I have the time or resources to continue developing and marketing the app—but I’ll try to polish it.

Key Takeaway and Outro

If I were to do this hackathon again, I would definitely bring in a third person focused purely on marketing, presentation, and business. With that, I truly think we would’ve won—and gotten more sleep.

Still, I’m very proud of what we accomplished with just two people.

If you made it this far, thanks for reading!

Details About the Infrastructure (For Those Interested)

Some of the cloud infrastructure we used:

  • Azure App Service

  • Azure SignalR Service

  • Azure Cosmos DB

  • Azure Functions

  • Azure AI Content Safety

The AI model used for room-tag questions was Claude Haiku. For context retrieval and storage, we planned to use Google Vertex AI RAG Engine, but due to technical difficulties, we used Cosmos DB again.

Ideally, the context storage would’ve been on Azure anyway, but with both people trying to work on the cloud simultaneously, we got tangled. My friend tried using GCP since he was more familiar with it.

There are definitely some technologies I’m missing here, but the major ones are listed.