← Om Badgujar
Project / 02
Multi-Agent · LangGraph
Multi-Agent Research Assistant (MARA)
A multi-agent research platform that automates web research, fact verification, summarization, and report generation. A stateful LangGraph workflow coordinates specialized LLM agents; Tavily fetches the web, Groq writes the report.
§ Problem
Why it exists.
Serious research means searching, cross-checking, summarising, and writing — four different jobs. One long prompt collapses them into a mediocre paragraph. Splitting the work across specialized agents keeps each step honest.
§ Approach
How it works.
- 01Modeled the pipeline as a stateful LangGraph workflow with dedicated agents for research, fact checking, summarization, and report generation.
- 02Integrated Tavily Search for real-time web retrieval so agents work from fresh sources, not stale training data.
- 03Used Groq LLM for low-latency, citation-aware summaries and structured final reports.
- 04Shipped FastAPI REST endpoints and a Streamlit frontend for interactive queries and downloadable reports.
§ Key features
What it does.
- Four specialized agents: Researcher, Fact-Checker, Summarizer, Report Writer.
- Live web retrieval via Tavily so answers reflect current sources, not stale training data.
- Structured, citation-aware reports downloadable from the Streamlit UI.
- Modular LangGraph — add or swap an agent without rewriting the pipeline.
§ Tech stack
Built with.
FastAPILangGraphLangChainGroq LLMTavily SearchStreamlit
- ◆End-to-end research reports with inline citations, not just a chat reply.
- ◆Fact-check step catches contradictions before they land in the summary.
- ◆Modular graph — swap or add an agent without rewriting the pipeline.