AI / ML research · 2026
Multi-Agent Dino Game Builder
An autonomous LangGraph multi-agent system (Architect, Engineer, QA, Scorer, Human-in-the-Loop) that generates a playable Chrome Dino–style game.
- LangGraph
- ReAct
- Databricks
- MLflow
The problem
Can a team of cooperating LLM agents autonomously build and refine a working game?
My role
Researcher & engineer (MS project).
What I built
A LangGraph multi-agent workflow (Architect, Engineer, QA, Scorer, Human-in-the-Loop) with conditional routing, persistent state, and memory checkpoints. Tool-using ReAct agents handle code generation, execution, debugging, fallback recovery, and QA of gameplay features, while MLflow tracks agent latency, groundedness, and reasoning quality.
Architecture & stack
- LangGraph orchestration: conditional routing, persistent state, checkpoints
- ReAct agents: codegen, execution, debugging, fallback recovery, QA
- Iterative feedback loops across development cycles
- LLMOps monitoring with MLflow (on Databricks)