⚡ Endee RAG Knowledge Base

A Retrieval-Augmented Generation (RAG) chatbot powered by Endee — a high-performance open-source vector database built for AI search.

Upload technical docs (PDFs, Markdown, Text), and the chatbot answers questions based only on those documents.

---

Problem Statement

Traditional keyword search fails when users ask conceptual questions like "How do I connect Prisma to MongoDB?". This project solves that with semantic search — it understands the meaning behind a query, retrieves the most relevant chunks from your documents stored in Endee, and feeds them to an LLM to generate a precise answer.

---

How It Works

The Python Workflow

Step	What Happens	Tool Used
1. Extract Text	Read PDFs, Markdown, and .txt files	PyMuPDF (fitz)
2. Chunking	Split text into ~500-character overlapping pieces	Custom chunker
3. Embeddings	Convert each chunk into a 384-dim vector	sentence-transformers/all-MiniLM-L6-v2
4. Store in Endee	Upsert vectors + metadata into the vector database	endee Python SDK
5. Retrieve	User question → vector → find top 3 closest chunks in Endee	Endee.query()
6. Generate	Send chunks + question to LLM → get a grounded answer	OpenAI GPT-4o-mini

# Embedding example (Step 3)
from sentence_transformers import SentenceTransformer
model = SentenceTransformer('all-MiniLM-L6-v2')
vector = model.encode("How to connect Prisma to MongoDB?")

Architecture Diagram

graph TD
    A["📄 PDFs / Markdown / Text"] --> B["Python: PyMuPDF Text Extraction"]
    B --> C["Chunking (500 chars, 50 overlap)"]
    C --> D["Sentence-Transformers Embedding"]
    D -->|384-dim vectors| E[("⚡ Endee Vector Database")]
    
    F["❓ User Question"] --> G["Sentence-Transformers Embedding"]
    G -->|Query vector| E
    E -->|"Top 3 context chunks"| H["Prompt Assembly"]
    F --> H
    
    H --> I["🤖 OpenAI GPT-4o-mini"]
    I --> J["✅ Final Answer"]

    style E fill:#1a1a2e,stroke:#e94560,stroke-width:2px,color:#fff
    style J fill:#0f3460,stroke:#16213e,stroke-width:2px,color:#fff

Loading

---

Quick Start (3 Commands)

Prerequisites

• Python 3.10+
• Docker installed and running

1. Install dependencies

git clone https://github.com/Vikash9546/Endee-RAG-Demo.git
cd Endee-RAG-Demo
python3 -m venv venv && source venv/bin/activate && pip install -r requirements.txt

2. Start Endee + ingest sample docs

docker run -d -p 8080:8080 --name endee endeeio/endee-server:latest && python ingest.py

3. Query the knowledge base

# Option A: Use Google Gemini (FREE — get key at https://aistudio.google.com/apikey)
export GEMINI_API_KEY="gemini_api_key"
python query.py "What is Endee built for?"

# Option B: Use OpenAI GPT-4o-mini (paid)
export OPENAI_API_KEY="sk-your-key-here"
python query.py "What is Endee built for?"

How it works: The script tries OpenAI first, then falls back to Gemini. If neither key is set, it still returns the raw semantic search results from Endee.

---

Option 3: "Ghost-Protocol" (Agentic AI Workflow)

An AI agent that can "browse" its own memory to complete a multi-step task for Automated Incident Response.

python incident_agent.py

This runs an autonomous loop where the agent:

1. Receives a server crash error.

⚡ Endee AI: Knowledge & Agentic Memory Platform

A high-performance implementation of RAG (Retrieval Augmented Generation) and Autonomous Agent Operations using the Endee Vector Database.

---

🚀 Core Features

1. 🤖 AI Knowledge Assistant (Stateful RAG)

A professional-grade conversational interface that allows you to chat with your private documents.

• Deep PDF/MD Ingestion: Parses and chunks documents into Endee.
• Stateful Memory: Remembers chat history for fluid, multi-turn follow-ups.
• Grounding & Citations: Answers are strictly based on provided context with automatic source attribution.
• Gemini 3-Flash Integration: Uses the latest fast-inference models for near-instant responses.

2. 🕵️ Agentic AI Memory (Ghost-Protocol)

An autonomous "Incident Response" agent that uses Endee as its long-term stateful memory.

• Autonomous Decision Engine: Detects the severity of incoming alerts.
• Memory-Driven Logic: Consults past "experience" in Endee to decide whether to Auto-Fix a server or Escalate to a human.
• Persistent State: Learns from every past incident logged in the vector store.

---

🛠️ Quick Start

1. Install Dependencies:

pip install -r requirements.txt

2. Configure Environment: Create a .env file with:

GEMINI_API_KEY=your_key_here

3. Launch the Dashboard:

streamlit run app.py

---

📁 Project Structure

• app.py: The main Streamlit dashboard featuring both AI modes.
• incident_agent.py: CLI logic for the Ghost-Protocol agent.
• ingest.py: Core logic for vectorizing text data into Endee.
• query.py: Script for direct semantic querying of the vector store.

---

© 2026 Endee AI Project. Built for high-speed intelligence.

---

How Endee Is Used

Endee is the core component of this application — it acts as the "memory" for the AI:

1. Index Creation — We create a vector index with dimension=384, space_type="cosine", and precision=FLOAT32 via client.create_index().
2. Upserting Vectors — Document chunks are embedded and stored with index.upsert(), including metadata (source file, chunk text, chunk index).
3. Querying — User questions are embedded with the same model and sent to index.query(vector=..., top_k=3) for nearest-neighbor retrieval.
4. RAG Pipeline — The top 3 matching chunks from Endee are injected into the LLM prompt as grounding context, ensuring the AI answers based on your documents.

---

Project Structure

endee-rag-demo/
├── app.py               # Streamlit chatbot UI (upload PDFs, ask questions)
├── ingest.py            # Ingestion pipeline: extract → chunk → embed → upsert
├── query.py             # CLI: semantic search + RAG generation
├── recommendation.py    # Bonus: product recommendation via vector similarity
├── agent.py             # Bonus: agentic AI memory pattern using Endee
├── data/                # Place your PDFs / Markdown / Text files here
│   └── endee_overview.md
├── docker-compose.yml   # One-command Endee server setup
├── requirements.txt     # Python dependencies
└── README.md            # This file

---

Example Output

╔══════════════════════════════════════════════════╗
║   Endee Knowledge Base — RAG Query Engine       ║
╚══════════════════════════════════════════════════╝

[1/3] Encoding question: "What is Endee built for?"
[2/3] Querying Endee for top 3 matching chunks...

  🔍  SEMANTIC SEARCH RESULTS (from Endee)
=======================================================

  [1] data/endee_overview.md  (distance: 0.4544)
      # Endee: High-Performance Vector Database ...

  [2] data/endee_overview.md  (distance: 0.4669)
      Endee is a specialized, open-source vector database ...

  🤖  RAG GENERATION (LLM + Endee Context)
=======================================================

  ╔══════════════════════════════════════════════╗
  ║  ✨  FINAL ANSWER                           ║
  ╚══════════════════════════════════════════════╝

  Endee is built for speed, efficiency, and scale required
  by production AI systems. It supports semantic search,
  RAG workflows, agentic memory, and hybrid search...

---

Tech Stack

Component	Technology
Vector Database	Endee
Embeddings	sentence-transformers/all-MiniLM-L6-v2
PDF Extraction	PyMuPDF
LLM	OpenAI gpt-4o-mini
Frontend	Streamlit
Containerization	Docker

---

License

MIT