AutoResearcher

Multi-Agent Research Crew for Instant, Citation-Rich Reports
Transforms a single prompt into a print-ready LaTeX PDF in under 10 minutes.

---

🚀 What AutoResearcher Produces

Give the system a prompt like:

“Create a 15–20 page report on AI agents in healthcare in 2025 for hospital CIOs.”

Within minutes you get:

• 50+ verified sources from Tavily
• Fact-checked claims by a dedicated agent
• Matplotlib charts (market forecast, risk matrix, trends)
• Structured academic writing (exec summary → conclusion)
• A complete PDF with title page, TOC, figures, citations
• No hallucinated references

⚡ End-to-end pipeline runs in <10 minutes
⚡ Zero manual formatting or citation work

---

❓ Why AutoResearcher Exists

Professionals in 2025 still lose 8–10 hours building a single research report:

• Searching and extracting sources
• Verifying claims
• Creating charts
• Writing structured academic sections
• Formatting citations
• Exporting to PDF or LaTeX

Traditional AI tools summarize text but cannot generate reproducible, citation-rich, multi-section research documents.
AutoResearcher solves this with a coordinated agent team.

---

🤖 How It Works: The Multi-Agent Crew

AutoResearcher uses a 7-agent hierarchical system built on the Vertex AI ADK for Python.

Agent	Model	Tool	Role
Supervisor	Gemini 2.5 Pro	—	Manages workflow + structured JSON task routing
Researcher	Gemini 2.5 Flash	Tavily	Collects real, live web sources
FactChecker	Gemini 2.5 Pro	Tavily	Cross-checks claims and stats
Visualizer	Gemini 2.5 Flash	Python REPL	Generates charts via Matplotlib
Writer	Gemini 2.5 Pro	—	Produces academic-grade sections
Formatter	Gemini 2.5 Flash	—	Builds final Markdown with TOC + references
Compiler	—	Pandoc + LaTeX	Converts Markdown → PDF

All agents share an in-notebook dictionary called MEMORY, enabling coordinated, stateful work.

System Architecture

---

▶️ Quickstart

Option A — Kaggle (Recommended)

1. Open the notebook
   https://www.kaggle.com/rayklanderman/autoresearcher-capstone
2. Add secrets:
   • PROJECT_ID
   • GEMINI_API_KEY
   • TAVILY_API_KEY
   • SERVICE_ACCOUNT_JSON
3. Run all cells

Option B — Google Colab

1. Open the Colab notebook
2. Add secrets (Runtime → Secrets) or edit code to input them manually
3. Install dependencies → run all cells

💡 Tip: Cell 7 displays the entire agent pipeline—perfect for demos.

---

🛠 Tech Stack

• Framework: Vertex AI ADK for Python
• Models: Gemini 2.5 Pro + Flash
• Tools: Tavily API, Matplotlib, Pandoc + LaTeX
• Runtime: Kaggle / Colab
• State Management: Shared in-notebook memory dictionary
• Security: All keys pulled from platform secrets

Install ADK:

pip install "git+https://github.com/google/adk-python.git@main"

---

📁 Repository Structure

CapstoneProject-Autoresearcher/
├── autoresearcher.ipynb          # Main Jupyter notebook
├── LICENSE                       # License file
├── README.md                     # This README
├── SECURITY.md                   # Security policy
├── CONTRIBUTING.md               # Contributing guidelines
├── CODE_OF_CONDUCT.md            # Community code of conduct
├── NOTICE                        # Legal notices and attributions
└── SECRET_HANDLING.md            # Secure handling of secrets & API keys

---

📄 License

For educational and capstone submission purposes only.
© 2025 Raymond Robert Klanderman