🔬 Autonomous Research Agent

AI agent that researches any topic — PubMed + Semantic Scholar for academic papers, DuckDuckGo for web. Multi-round autonomous research with structured Markdown reports.

Built with Python + GLM-5.2 LLM. Part of my AI Agent Career Portfolio.

---

🎯 Features

• 🔬 Scholarly + Web Research — Auto-detects academic topics → uses PubMed & Semantic Scholar. General topics → DuckDuckGo web search.
• 🔄 Multi-round agent loop — Generates diverse queries → searches → extracts → synthesizes → refines based on gaps
• 📊 Structured reports — Executive Summary, Key Findings, Detailed Analysis, Controversies, Practical Implications, Sources
• 🎨 Beautiful CLI — Rich formatting with live progress panels
• ⚙️ Configurable — Rounds, max results, temperature, sources

---

🏗️ Architecture

Topic → [LLM: Generate queries] → [Is it academic?]
                                        ├─ Yes → PubMed + Semantic Scholar
                                        └─ No  → DuckDuckGo web search
                                              ↓
                                    [Extract abstracts/content]
                                              ↓
                                    [LLM: Synthesize round]
                                              ↓
                                    [Refine queries from gaps]
                                              ↓
                                    [Generate structured report]

---

🚀 Quick Start

git clone https://github.com/Meshkat22/research-agent.git
cd research-agent
pip install -r requirements.txt

# Academic research (auto-uses PubMed + Semantic Scholar)
python main.py -t "BCI motor imagery CNN Transformer hybrid architectures" -r 1

# General research (auto-uses DuckDuckGo)
python main.py -t "AI agent job market trends 2026" -r 2 -o report.md

# Verbose debug mode
python main.py -t "Quantum computing 2026" --verbose

---

🛠️ Tech Stack

Layer	Technology
LLM	GLM-5.2 (OpenAI-compatible)
Academic Search	PubMed Entrez API + Semantic Scholar
Web Search	DuckDuckGo
Content Extraction	trafilatura + BeautifulSoup fallback
CLI	rich
Language	Python 3.9+

---

📁 Project Structure

research_agent/
├── main.py              # CLI entry point
├── agent.py             # ResearchAgent class (orchestration)
├── scholarly_tools.py   # PubMed + Semantic Scholar integration
├── tools.py             # DuckDuckGo web search + scraping
├── config.py            # LLM, agent, PubMed configuration
├── requirements.txt     # Dependencies
└── README.md

---

🎯 Impact

This project demonstrates:

1. Agent Loop Design — Multi-step autonomous research cycle
2. Tool Selection — Auto-detects academic vs web-appropriate sources
3. API Integration — PubMed, Semantic Scholar, DuckDuckGo, OpenAI-compatible
4. Prompt Engineering — Query generation, synthesis, structured reporting
5. Error Resilience — Graceful degradation when APIs fail (LLM knowledge fallback)
6. Clean Architecture — Separated concerns, configurable, testable

---

🔜 Roadmap

• Streamlit web UI
• SerpAPI fallback for web search
• Europe PMC / OpenAlex integration
• PDF report output
• LangFuse observability
• Source claim verification

---

Built by Meshkat Ahmad — AI Agent Career Portfolio Project 1/7