Multi-Agent Research Assistant System

An autonomous LLM-powered agent team designed to extract, cluster, and reason over scientific papers. It generates structured summaries, identifies thematic contradictions, and proposes follow-up experiments. The deployed version of this app is available here (until my aws credits run out): http://3.16.161.72:8501/

Project Goals

• Automate literature review workflows
• Cross-reference research papers by methods, outcomes, and contradictions
• Propose follow-up experiments using LLM reasoning

System Overview

This system uses a team of large language model agents:

• ExtractorAgent: summarizes papers requested from semantic scholar API (I requested my own API key but publics keys are available)
• ClusteringAgent: identifies topic clusters and conflicting findings
• HypothesisAgent: generates new questions or experimental ideas

Each agent runs independently and collaborates via a central controller (custom orchestration).

Technologies

• LLM Backend: OpenAI GPT-4
• Orchestration: Custom
• Memory Store: MongoDB
• Containerization: Docker-compose
• Backend: FastAPI, pydantic (schema validation), aioredis (rate limiting)
• Frontend: Streamlit
• Auth: Firebase
• Deployment: AWS EC2

RUN

• If you are using vscode, reopen the directory using the provided .devcontainer.
• Add your API keys to your .env (example is included)
• To run API uvicorn app.api:app --host 0.0.0.0 --port 8000
• To launch streamlit cd frontend then streamlit run streamlit_app.py then navigate to http://localhost:8501/ or whichever is specified

OR

• To run via your terminal, run main.py.