ATG Technical Assignment: Multi-Agent Debate DAG using LangGraph Machine Learning Intern Deliverables Task Overview Construct a debate simulation system using LangGraph where two AI agents engage in a structured argument over a fixed topic. The system should use memory, control over turns, and a judging node that evaluates the debate logically and declares a winner. Deadline: 48 hours from task assignment Objective Design and implement a LangGraph-based workflow with two alternating agents debating a fixed topic. The system should preserve and summarize the dialogue history, validate logical progression, and end with an automated judgment using a dedicated node. Workflow Nodes •UserInputNode: Accepts the debate topic at runtime from the user. •AgentA and AgentB: Two profession-specific personas (e.g., Scientist vs Philoso- pher) that take alternate turns making arguments. •Rounds: The debate must consist of exactly 8 rounds — 4 arguments per agent, alternating. •MemoryNode: Stores and updates a structured summary or full transcript of argu- ments. Each agent only receives relevant memory (no full state sharing). •JudgeNode: – Reviews memory and all argument nodes – Produces a full debate summary – Declares a winner with logical justification 1 Additional Requirements •All node messages and responses must be logged to a file, including the final judgment. •Implement state validation to ensure: – Each agent only speaks in their assigned turn – No argument is repeated – Logical coherence is maintained across the flow •The entire debate must operate via a clean CLI interface. •Include a DAG diagram (either static using graphviz, or generated programmatically via LangGraph tools). •Submit a full log of all state transitions, memory content, and final verdict. Expected Output •CLI-based interaction: Enter topic for debate: Should AI be regulated like medicine? Starting debate between Scientist and Philosopher... [Round 1] Scientist: AI must be regulated due to high-risk applications. [Round 2] Philosopher: Regulation could stifle philosophical progress and autonomy. ... [Round 8] Philosopher: History shows overregulation often delays societal evolution. [Judge] Summary of debate: ... [Judge] Winner: Scientist Reason: Presented more grounded, risk-based arguments aligned with public safety principles. Deliverables
- Source Code: Modular files for all LangGraph node definitions and execution logic.
- README.md: •How to run the program and install dependencies •Node and DAG structure explained 2
- DAG Diagram: Visual layout of the LangGraph architecture (static image or auto- generated).
- Chat Log File: Full log of all messages, transitions, memory updates, and final verdict.
- Demo Video: 2–4 minute walkthrough (screen + face-cam) explaining: •Project structure and key files •CLI flow •Judgment process Submission Format •GitHub repository or zipped folder containing: – All source code – README.md – DAG diagram – Log file – Demo video (or shareable link) Evaluation Criteria •Correctness and functionality of LangGraph DAG •Debate round control and memory handling •Judge logic and decision explanation •Code quality and documentation •Communication and clarity in the demo video