How mihirkothari04/ai-research-assistant-network works
Project Type and Purpose
• This is a full-stack, containerized, multi-agent AI research system (“Research Assistant Network”).
• Purpose: accept an open-ended research question via a web UI, then execute a directed pipeline of five specialized AI agents—InfoGatherer, InsightGenerator, DataAnalyst, HypothesisTester, ReportCompiler—to autonomously collect data, analyze it, test hypotheses, and produce a report.
File Structure and README
• Top-level scripts: main.py, run_server.py (entry points), Dockerfile, docker-compose.yml (deployment).
• src/
• api/ (Flask REST/web layer + templates)
• agents/ (five agent modules + a BaseAgent)
• config/ (application configuration)
• orchestration.py (LangChain/LangGraph orchestration graph)
• Logs (flask.log, flask_error.log), requirements.txt, DEPLOYMENT.md, and environment config via .env.
Main Components
a. Client/Web browser
b. API Layer
• Flask application (src/api/api.py + index.html template)
c. Orchestration Engine
• LangChain/LangGraph directed graph (src/orchestration.py)
d. Agent Modules (five nodes in the graph)
1. InfoGathererAgent
2. InsightGeneratorAgent
3. DataAnalystAgent
4. HypothesisTesterAgent
5. ReportCompilerAgent
e. External Services
• OpenAI GPT-4 (via OpenAI API)
• Academic data sources (Semantic Scholar, arXiv, CrossRef, Bing Search)
f. Data Stores
• Vector Database (ChromaDB for embeddings in storage/vector_db/)
• File storage (storage/research_tasks/, storage/reports/)
• Logging (flask.log)
g. Containerization/Deployment
• Docker, Docker Compose
Relationships and Data Flows
• User → HTTP Request → Flask API
• Flask API → passes query into Orchestration Engine
• Orchestration Engine → triggers InfoGathererAgent → fetches documents from academic APIs → stores into ChromaDB → returns data to engine
• Engine → invokes InsightGeneratorAgent → summarizes and passes output downstream
• Engine → invokes DataAnalystAgent → processes data, generates visuals
• Engine → invokes HypothesisTesterAgent → proposes/tests hypotheses
• Engine → invokes ReportCompilerAgent → builds final report (Jupyter/PDF/Slides) → writes to storage/reports/
• Final Report → Flask API → user downloads via web UI
Architectural Patterns and Design Principles
• Pipeline/Micro-workflow: Directed Graph orchestration (LangGraph) defines ordering and data dependencies among agents.
• Modular agents conforming to BaseAgent interface for easy extension.
• Clean separation: API layer, orchestration logic, agent implementation, storage, config.
• Containerized deployment ensures environment consistency.
Technologies, Frameworks, and Libraries
• Python 3.11
• Flask (REST API + web UI)
• LangChain & LangGraph (orchestration)
• OpenAI API for LLM reasoning
• ChromaDB (vector store)
• Docker & Docker Compose
• Jupyter/PDF/Slides for report outputs
Final Advice
• Be exhaustive: include entry points (main.py/run_server.py), config (config.py), deployment (Docker).
• Directional dependencies: which agent feeds data to which.
• Both synchronous flows (web request → orchestrator → report) and asynchronous or storage interactions (vector DB reads/writes).
• Iterate: validate with a peer or the person who implemented orchestration.py to ensure agent boundaries and data contracts are accurately reflected.