GraphRAG Inference Hackathon by TigerGraph

Welcome to the GraphRAG Inference Hackathon by TigerGraph!

A beginner-friendly online coding challenge where you'll prove that graphs 
make LLM inference faster, cheaper, and smarter than vector-based RAG alone.

About the Opportunity:

LLMs are spreading rapidly across every industry, and as they do, token 
consumption is exploding. Companies pay more, wait longer, and hit context 
limits faster every quarter. This isn't an academic exercise. It's a real, 
growing pain in production AI today.

The GraphRAG Inference Hackathon by TigerGraph is your chance to prove that 
graphs solve this. You'll build three pipelines side-by-side (raw LLM, Basic 
RAG, and GraphRAG) and let the numbers tell the story.

The headline metric: token reduction with maintained accuracy. Show us how 
much GraphRAG cuts down without dropping answer quality, and you'll have 
built the kind of project that gets hiring managers, AI teams, and the 
open-source community paying attention.

New to GraphRAG? No problem. This hackathon is designed to be 
beginner-friendly. 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. We provide documentation, starter resources, and 1:1 
mentoring sessions for the Top 10 teams.

Whether you're a student exploring GenAI, a developer curious about graphs, 
or an AI engineer building production RAG systems, this hackathon is your 
playground to learn, build, and compete.

What You will Build:

Three pipelines that answer the same questions on the same data, plus an 
interactive comparison dashboard.

• Pipeline 1: LLM-Only (Baseline). A prompt goes in, an output comes out. 
  No retrieval. Your worst-case baseline.
• Pipeline 2: Basic RAG (Vector + LLM). Vector embeddings retrieve similar 
  chunks, the LLM generates an answer using that context. The current 
  industry standard.
• Pipeline 3: GraphRAG (Graph + LLM). Built on top of the TigerGraph 
  GraphRAG repo. TigerGraph extracts entities, maps relationships, and 
  performs multi-hop reasoning. The LLM generates the final answer using 
  this filtered, structured context, not a giant document dump.
• Comparison Dashboard: An interactive page where you enter one query and 
  all three pipelines respond side-by-side. Tracks tokens used, response 
  latency, cost per query, and answer accuracy (evaluated using LLM-as-a-Judge 
  + BERTScore via Hugging Face). One query in, three answers + metrics out, 
  side-by-side. That's the story.

Full Problem Statement & Hackathon Guidebook:
Hackathon Problem Statement

This Notion guide has everything you need: the two-round structure, dataset 
requirements, judging criteria, deliverables, accuracy evaluation methodology, 
mentoring details, and step-by-step setup. Bookmark it before you start 
building.

Join the Hackathon WhatsApp Group:
Hackathon Participants WhatsApp Group

This is where the action is. Get help from mentors, connect with other 
builders, ask questions, share progress, and stay updated on announcements 
throughout the hackathon.

Eligibility and Registration:

• Open to students and working professionals globally. Beginners are welcome!
• Team size: 1 to 5 members (solo participation allowed).
• Each participant can be part of only one team.
• Cross-institutional teams are welcome.
• Participation is completely free. 

Rewards and Prizes:

• Winner (1st Place): $250 cash prize, certificate, and recognition for 
  building the best GraphRAG inference system.
• 1st Runner-Up (2nd Place): $150 cash prize and certificate for outstanding 
  GraphRAG implementation.
• 2nd Runner-Up (3rd Place): $100 cash prize and certificate for strong 
  graph utilization and efficient inference design.
• TigerGraph Community Leads' Exclusive Award: $100 cash prize and 
  certificate. A special award for creativity, innovation, and community 
  spirit.

• $100 Content Bounty: Want to earn extra by sharing your hackathon journey?
  Create content (blog posts, social media posts, video walkthroughs, tutorials)
  about your GraphRAG build and stand a chance to win a $100 content bounty.

  Sign up here