Everything you want
to ask before you register.

The hackathon is open to students and working professionals globally — beginners welcome. You don't need prior experience with TigerGraph or graph databases. If you know Python and have played with LLMs or APIs, you're ready to build.

1 to 5 members per team. Solo participation is allowed. Each participant can only be part of one team. Cross-institutional teams are welcome.

No. Participation is completely free.

Register solo on Unstop and join the official WhatsApp group. There's a dedicated channel where participants post skills and look for teammates.

May 10, 2026 · 11:59 PM IST. Building phase opens May 4 and runs through May 16. See the full timeline.

Three pipelines that answer the same questions on the same data: (1) LLM-Only (no retrieval), (2) Basic RAG (vector embeddings + LLM), and (3) GraphRAG built on the TigerGraph GraphRAG repo. Plus an interactive comparison dashboard: enter one query, run it through all three, display side-by-side answers with tokens, latency, cost, and accuracy. For Pipeline 3 you can use the GraphRAG repo as-is or customize it (tune retrievers, modify prompts, extend the schema) — both paths are valid; we judge on outcomes.

No. TigerGraph is providing documentation, starter resources, and mentoring sessions for the Top 10 teams. The hackathon is designed to be beginner-friendly.

Any LLM you want — Gemini free tier, OpenAI, Anthropic, open-source models via Hugging Face, local Ollama setups. Most major providers have free tiers that are more than enough for hackathon-scale usage. Use the same model across all three pipelines for a fair comparison.

Bring your own — legal, medical, news, support tickets, scientific papers, podcast transcripts, anything that excites you. The best datasets have lots of natural connections between people, places, events, and topics. Make sure the content is public domain or properly licensed. Round 1 minimum: 2 million tokens of text. Top 10 teams scale to 50–100M tokens in Round 2.

Yes — for Round 2. Top 10 teams receive $50 worth of Gemini API credits per team, shared via secure endpoints. No out-of-pocket cost during Round 2. For Round 1, free tiers from Gemini, Ollama (local), and similar are more than enough.

Yes — your GraphRAG pipeline must be built on TigerGraph's official repo: github.com/tigergraph/graphrag. Two valid paths: Path A — use the repo as-is via Docker, ingest your dataset, hit its REST/Python APIs (best if you want to focus on benchmarking + storytelling). Path B — customize: tune retriever parameters (top_k, num_hops, community_level), swap retrievers (Hybrid, Community, Sibling), modify prompts, extend the schema. Both paths are judged on outcomes — % improvement and answer accuracy — not on which path you took.

Two complementary Hugging Face approaches: LLM-as-a-Judge (a free hosted model grades each answer PASS/FAIL) and BERTScore (semantic similarity to the correct answer). Bonus points kick in at LLM-as-a-Judge ≥ 90% pass rate and BERTScore F1 rescaled ≥ 0.55 (or raw ≥ 0.88). Token reduction only counts if your GraphRAG accuracy holds vs your Basic RAG baseline.

All deliverables go through Unstop before May 16, 2026 · 11:59 PM IST. Detailed submission guidelines (required files, demo video length, dashboard format) will be shared with all registered teams.

Eight deliverables, all uploaded via Unstop: architecture diagram, comparison dashboard, benchmark report (tokens / cost / latency / accuracy per pipeline), 5–7 minute demo video, public GitHub repo (built on the TigerGraph GraphRAG repo), blog post (Medium / Hashnode / Dev.to), social media post tagged @TigerGraph with #GraphRAGInferenceHackathon, and a product feedback interview (Top 5–10 only). Full list on the Prizes page.

No. The deadline is hard. Unstop closes submissions automatically at 11:59 PM IST on May 16.

Four weighted criteria, totalling 100%: Token Reduction (30%), Answer Accuracy (30%), Performance (20%), and Engineering & Storytelling (20%). Bonus points for high accuracy: LLM-as-a-Judge pass rate ≥ 90% and BERTScore F1 rescaled ≥ 0.55 (or raw ≥ 0.88). Token reduction only counts if accuracy is maintained or improved vs Basic RAG. Full breakdown on the Prizes page.

Top 10 are announced on May 18 and advance to Round 2: scale your winning approach to 50–100M tokens with $50 of Gemini API credits per team (provided by TigerGraph). 1:1 mentoring with TigerGraph engineers runs May 18–24, live presentations are on May 25, and winners are announced May 26. Final benchmarks may be featured in the official TigerGraph blog and case studies.

Cash prizes are in USD and will be transferred to the team's designated account after the winner announcement on May 26. Certificates are delivered digitally within 30 days.

Email devanshu.saxena@tigergraph.com, call +91 74043 13376, or post in the WhatsApp group. Our 20 community leads also help answer questions during the build phase.

Yes. Live workshops on GraphRAG fundamentals and TigerGraph basics will be scheduled after registration closes. Details will be posted in the WhatsApp group and via Unstop.

Your code stays yours. We only ask that the specific GraphRAG benchmark implementation was built during this hackathon's build window (May 4 – May 16, 2026). Prior libraries you authored are fine.