Win from a $700 prize pool across five tracks, plus TigerGraph recognition and certificates for all participants.
Awarded to the team that builds the most compelling, production-ready three-pipeline benchmark — biggest verified token reduction, accuracy held or improved, and the clearest story told with their numbers.
Strong execution across all three pipelines with clear, credible benchmarks and accuracy that holds vs Basic RAG.
Strong graph utilization and efficient inference design — best use of multi-hop reasoning to keep tokens honest while accuracy holds.
A special award picked by TigerGraph's community leads — for the team that surprises us with the most creative angle, or who shows up for the community along the way.
Earn extra by sharing your hackathon journey. Create blog posts, social posts, video walkthroughs, or tutorials about your GraphRAG build and stand a chance to win a $100 bounty. Tag #GraphRAGInferenceHackathon and @TigerGraph. Sign up separately on Luma.
All Top 10 teams receive appreciation certificates from TigerGraph.
Every participant who submits a valid solution receives a digital participation certificate.
Four weighted criteria, totalling 100%. Token reduction only counts if accuracy holds — a 70% token cut with 20% accuracy loss is a regression, not a win.
| Criteria | Weight | What we're looking for |
|---|---|---|
| Token Reduction | 30% | % improvement in tokens and cost per query vs Basic RAG. Show us the numbers. |
| Answer Accuracy | 30% | Quality maintained or improved vs Basic RAG. Evaluated with LLM-as-a-Judge + BERTScore. |
| Performance | 20% | Latency, throughput, and overall system efficiency. A fast pipeline matters as much as a cheap one. |
| Engineering & Storytelling | 20% | Clean architecture, working dashboard, clear demo video, blog post that tells the story. |
Strong accuracy is heavily rewarded. Submissions hitting these bars earn extra points: LLM-as-a-Judge pass rate ≥ 90% and BERTScore F1 rescaled ≥ 0.55 (or raw ≥ 0.88). Hitting both unlocks the maximum bonus.
TigerGraph's official GraphRAG reference. Use it as your starting point for graph construction, entity + community identification, and multi-hop reasoning patterns. Clone, extend, benchmark.