alketbak/llm-research-agent

title: LLM Research Agent emoji: 🔍 colorFrom: green colorTo: blue sdk: docker pinned: false ---

LLM Research Agent

An agentic research assistant that replaces hours of manual web research with a structured, cited report in seconds — powered by LangGraph, Groq (Llama 3.1), and Tavily.

The problem

Google gives you links. You still have to open each one, read it, extract the relevant parts, and write a summary yourself — that takes 30–60 minutes.

This agent searches multiple sources simultaneously, reads them, extracts key information, and returns a structured report with cited sources in under 30 seconds.

Developed as a side project to automate the literature research process during my work on Predicting Startup Success Using LLMs — a research paper on in-context learning published on arXiv."

🔴 Live Demo

How it works

The agent uses a 3-node LangGraph pipeline: generates search queries → searches the web in parallel via Tavily → synthesises findings into a structured report with cited sources.

It also supports multi-turn conversations with persistent memory via LangGraph MemorySaver — ask follow-up questions and it remembers the context.

API Endpoints

Local setup

git clone https://github.com/alketbak/llm-research-agent.git
cd llm-research-agent
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
echo "GROQ_API_KEY=your-key-here" >> .env
echo "TAVILY_API_KEY=your-key-here" >> .env
uvicorn main:app --reload

Tech stack

• —LangGraph — agentic orchestration with conversation memory
• —Groq — ultra-fast inference with Llama 3.1 8B
• —Tavily — real-time web search API
• —FastAPI — REST API framework
• —Docker — containerised deployment