HUNT aggregates fragmented contest sources, scores expected value under deadline pressure, and answers natural-language questions about what to enter next — with evidence. For this hackathon, Ask/explain paths call NVIDIA Nemotron via Nebius Token Factory when NEBIUS_API_KEY is set.
How we built it
- Ranking + Why engine in TypeScript (
src/lib/contests.ts,src/lib/hunter.ts) - Nebius client (
src/lib/nebius.ts) → Token Factory chat completions - Next.js board + Ask UI + APIs
- MIT license at repo root (required)
Challenges
Keeping deadline-critical hunts above larger-but-later prize pools; making Why answers cite real sources instead of vibes.
Accomplishments
Working demo board, explainable ranking, Nebius/Nemotron hook, dual-track contest strategy toward ₩100M goal.
What we learned
Prize hunting is an agentic ops problem: fragmented intel + time decay + fit beats raw prize size.
What's next
Live ingest scrapers, more FX-aware EV, Token Factory for multi-step hunt plans, Tavily for fresh listing discovery (Best Use of Tavily eligible).
Built With
- nebius
- next.js
- nvidia
- tailwind
- typescript
