🤖 Agentic RAG with LlamaIndex & Gemini

A multi-document reasoning chatbot that uses an autonomous ReAct agent to cross-reference technical documentation in real time. Built with LlamaIndex, Google Gemini, and Streamlit.

---

What It Does

Instead of running a single vector search per query, this project implements a full agentic loop: the agent decides which documents to query, calls the appropriate tools, synthesizes information across sources, and produces a grounded answer — all autonomously.

Demo scenario: The agent can cross-reference n8n automation documentation against Claude Code technical docs to answer comparative questions like "How does n8n's trigger-action model compare to Claude Code's agentic loop?"

---

Architecture

User Query
    │
    ▼
ReActAgent (LlamaIndex 0.14 Workflow-based)
    │
    ├── n8n_documentation_tool      ──► VectorStoreIndex (storage/n8n)
    └── claude_code_documentation_tool ──► VectorStoreIndex (storage/claude)
                                              │
                                         PyMuPDF + GoogleGenAI Embeddings
                                         (gemini-embedding-001)

The agent uses Gemini 2.5 Flash as its reasoning backbone and gemini-embedding-001 for semantic vector search.

---

Stack

Layer	Technology
Agent framework	LlamaIndex Core 0.14 (ReActAgent)
LLM	Google Gemini 2.5 Flash
Embeddings	Google Gemini Embedding 001
Vector store	LlamaIndex in-memory + disk persistence
PDF parsing	PyMuPDF (fitz)
UI	Streamlit

---

Project Structure

├── app.py          # Streamlit UI + agent initialization
├── ingest.py       # Document ingestion pipeline (PDF → vector index)
├── docs/
│   ├── n8n/        # Drop your n8n documentation files here
│   └── claude/     # Drop your Claude Code documentation files here
├── storage/        # Persisted vector indices (git-ignored)
├── requirements.txt
└── .env            # GOOGLE_API_KEY (git-ignored)

---

Setup

1. Clone and create a virtual environment

git clone https://github.com/mfpNahu07/Agentic-RAG-LlamaIndex.git
cd Agentic-RAG-LlamaIndex
python -m venv .venv
.venv\Scripts\Activate.ps1   # Windows
# source .venv/bin/activate  # macOS/Linux

2. Install dependencies

pip install -r requirements.txt

3. Configure your API key

Create a .env file in the project root:

GOOGLE_API_KEY=your_api_key_here

4. Add your documents

Drop your files into docs/n8n/ and docs/claude/. Both .txt and .pdf formats are supported.

5. Build the vector indices

python ingest.py

6. Run the app

python -m streamlit run app.py

---

How to Use

Ask any question that requires reasoning across both documents. Examples:

• "What is the agentic loop in Claude Code and what are its steps?"
• "Compare the automation approach of n8n with Claude Code's tool execution model."
• "Which platform gives more control to the user when executing actions, and why?"

The agent will call one or both documentation tools, retrieve the relevant chunks, and synthesize a grounded answer.

---

Adding More Documents

To index additional sources:

1. Create a new folder under docs/ (e.g., docs/langchain/)
2. Add a new QueryEngineTool in app.py pointing to the new index
3. Add a new indexing block in ingest.py
4. Re-run python ingest.py

---

License

MIT