🔬 Autonomous Research Agent

An AI agent that autonomously searches the web, reads sources, and generates full research reports — powered by LangChain, Groq LLaMA-3, and Tavily Search.

---

What Is An AI Agent?

An AI agent is a program that can decide what to do next on its own, rather than following a fixed script. It loops through thinking, acting, and observing results until it has enough information to answer. Unlike a simple chatbot that answers from memory, an agent actively goes out and finds information in real time. This one searches the web, reads webpages, and synthesises everything into a structured report — all without human guidance.

---

How It Works

The agent follows the ReAct (Reasoning + Acting) loop:

┌─────────────────────────────────────────────────────────┐
│                     ReAct Loop                          │
│                                                         │
│   User Topic                                            │
│       │                                                 │
│       ▼                                                 │
│   💭 Think ──► What should I search for?               │
│       │                                                 │
│       ▼                                                 │
│   🔍 Search ──► Tavily queries live web                 │
│       │                                                 │
│       ▼                                                 │
│   📄 Read ──► Fetch full page content                   │
│       │                                                 │
│       ▼                                                 │
│   💭 Think ──► Do I have enough? What's missing?       │
│       │                                                 │
│       ├─── No ──► 🔍 Search again (new angle)          │
│       │                                                 │
│       └─── Yes ──► ✅ Generate final report             │
└─────────────────────────────────────────────────────────┘

Each loop: Think → Search → Read → Think → Search → ... → Answer

---

Features

• 🤖 Autonomous multi-step research — the agent decides what to search and when to stop
• ⚡ Powered by Groq — blazing fast LLaMA-3 inference (free tier)
• 🌐 Real-time web search — Tavily finds current, accurate sources
• 📖 Deep reading — agent fetches and reads full webpage content
• 🎬 Live streaming UI — watch the agent think and act in real time
• 📄 Structured reports — summary, key findings, analysis, citations
• 💾 Export to PDF and TXT — download professional reports instantly
• 🔒 100% free APIs — no OpenAI, no paid services

---

Quick Start

1. Get your free API keys

Service	URL	Free Tier
Groq (LLM)	https://console.groq.com	Unlimited (rate-limited)
Tavily (Search)	https://tavily.com	1000 searches/month

2. Clone and install

git clone <your-repo-url>
cd research-agent
pip install -r requirements.txt

3. Configure environment (optional)

cp .env.example .env
# Edit .env and add your API keys

Or enter them directly in the app sidebar — no .env file required.

4. Run the app

streamlit run app.py

Open your browser to http://localhost:8501

---

Example Research Topics

Try these prompts for impressive demos:

1. "Latest breakthroughs in fusion energy and nuclear power 2024"
   → Pulls recent ITER/NIF news, power generation data, timeline predictions

2. "How is AI changing drug discovery and pharmaceutical research?"
   → AlphaFold, ML molecular design, FDA approvals, company case studies

3. "State of electric vehicle adoption globally in 2024"
   → Sales figures, country comparisons, infrastructure stats, manufacturer data

4. "Best practices for building production-grade RAG systems"
   → Chunking strategies, embedding models, retrieval techniques, benchmarks

5. "Impact of the 2024 US election on global markets and trade policy"
   → Economic analysis, sector impacts, expert predictions, historical parallels

---

Architecture

ReAct Pattern

ReAct (Yao et al., 2022) combines Reasoning and Acting in an interleaved loop. At each step the LLM:

1. Emits a Thought — natural language reasoning about the situation
2. Selects an Action — which tool to call and with what input
3. Receives an Observation — the tool's return value
4. Uses the observation to inform the next thought

This lets the model correct itself mid-task — if a search returns poor results, it immediately rethinks the query.

Component Map

app.py              ← Streamlit UI, live streaming, download buttons
agent.py            ← ReAct AgentExecutor (LangChain + Groq LLaMA-3)
tools/
  search.py         ← Tavily web search → formatted results
  reader.py         ← requests + BeautifulSoup → clean text
report_gen.py       ← section parser, Markdown formatter, FPDF2 PDF generator

---

Demo

[Add a GIF or screenshot of the app here]

---

Project Structure

research-agent/
├── app.py              # Main Streamlit application
├── agent.py            # LangChain ReAct agent
├── report_gen.py       # Report parsing and PDF generation
├── tools/
│   ├── __init__.py
│   ├── search.py       # Tavily search tool
│   └── reader.py       # URL content reader tool
├── requirements.txt
├── .env.example
└── README.md

---

License

MIT © 2024 — free to use, modify, and distribute.

---

Day 7 of a 7-day AI portfolio sprint. Built with LangChain, Groq, Tavily, and Streamlit.