Most hackathons ask you to build a demo. DOGFOOD 2026 asked me to build the thing that would score me.
Hackathon Raptors has run 35 hackathons across 85+ countries. This time they gave every team the same spec and one job: build the submission-and-judging platform they would actually run. No tracks. 72 hours. The winner gets forked and used for real events.
I entered with JudgeForge ("Hack The Dog"). Here's what I think DOGFOOD really measures, and the decisions I made because of it.
The real question isn't "what can you build?"
When every team builds the same product from the same spec, features stop being the differentiator. Everyone ships events, teams, submissions and scoring. What's left is judgment:
Could an organizer run this tomorrow? If it doesn't start on a laptop with one command, nothing else matters.
Do you claim only what you can prove? The spec is a ladder of tiers (T1-T4) checked by an acceptance harness.
Did you cut scope on purpose? The organizers' own intro warns that trying to build everything is the quickest route to something mediocre.
So I stopped asking "what else can I add?" and started asking "what would I be comfortable maintaining?"
Decision #1: no database
JudgeForge is a Spring Boot backend and a React/Vite frontend served by Nginx. Persistence is plain JSON files.
That isn't laziness, it's an adoption argument. A platform that needs a database running, migrated and seeded has three ways to fail before the first login. A JSON file has one.
What I gained What I paid
Zero setup: docker compose up and it's live No concurrency guarantees at scale
State I can open, read and diff in a text editor No ad-hoc queries
Nothing to migrate between versions A ceiling on event size
I made the trade knowingly. A hackathon-sized event sits well below that ceiling, and the storage layer is the first thing I'd put behind an interface the day a real event outgrows it.
Decision #2: claim honestly
My .dogfood.toml claims T1 through T4. The acceptance script verifies T1 and T2, and all 7 checks pass. T3 and T4 aren't covered by that script, so I verified them by hand and documented exactly how.
I'd rather show my work than let a green checkmark imply something it doesn't. A judging platform is a trust product. If I'm vague about what my own project proves, why would anyone trust the scores it produces?
What's in the box
Organizers create events. Participants form teams. The whole stack, backend, frontend and Nginx, comes up with Docker Compose on localhost:8080, under an MIT license.
The hardest problem wasn't a feature. It was proof.
The hardest part of this build wasn't adding another feature. It was proving the platform behaved the way the hackathon required.
I started with a mistake: I read "offline" too literally. I spent time worrying that the very first docker compose up had to work with no network at all. Then I checked the organizer clarification, and the real requirement was different: the application must run without external services once the Docker images are prepared. That's the requirement that matters for a self-hosted platform, so that's the one I tested:
- Make a completely fresh clone.
- Build the containers from scratch.
- Bring the stack up, then shut it down.
- Disable the network.
- Restart with
docker compose up -d --no-build.
The clean-clone run also exposed something I hadn't considered. My Compose file used fixed container_name values, and on the same machine a fresh clone collided with the containers already there. No amount of feature work would have caught that.
It taught me that "works on my machine" isn't enough for a self-hosted submission. The deployment path itself has to be tested from a clean environment.
Try it-
bash
git clone https://github.com/rishikeshdas887/judgeforge-dogfood-2026
cd judgeforge-dogfood-2026
docker compose up
open http://localhost:8080
Repo: https://github.com/rishikeshdas887/judgeforge-dogfood-2026
Demo video: https://www.youtube.com/watch?v=vJxb4QcTrbo
The takeaway -
Dogfooding means eating what you cook. JudgeForge is small on purpose, and every claim in its README is one I can back up. That's the platform I'd want judging me.
Built for DOGFOOD 2026 by Hackathon Raptors. #Dogfood2026
Top comments (0)