Cerevello : AI-Powered Real-Time RAG System with Ticketing and Multimodal Support

This project is a real-time Retrieval-Augmented Generation (RAG) system, powered by OpenAI's LLMs, integrated with a FastAPI backend and a multimodal ingestion pipeline. It enables users to interact with enterprise knowledge through document upload, querying, and support ticketing, all while leveraging vector search via Qdrant.

---

🧠 Key Features

• RAG (Retrieval-Augmented Generation) using OpenAI and Qdrant
• Multimodal ingestion pipeline: Supports .pdf, .docx, .png
• OpenAI Agent + Tool integration to dynamically call backend services
• FastAPI microservices for RAG & ticketing
• Support ticketing system with MongoDB
• Semantic vector search using OpenAI embeddings
• Client endpoints for interacting with the system in real-time

---

🧱 Architecture Overview

This system is composed of several interconnected services:

1. 📦 RAG API (FastAPI Server)

Endpoints to:

• /query: Search and retrieve knowledge with OpenAI + Qdrant
• /upload: Bulk upload documents
• /update: Update existing vector entries
• /delete: Delete documents
• /llm-query: Basic LLM RAG
• /mcp/query: OpenAI Agent-based query routing with tools

2. 🧾 Ticketing API (FastAPI Microservice)

Endpoints to:

• /create-ticket: Submit a new support request
• /tickets: View all tickets
• /tickets/{userId}: View user-specific tickets

3. 🧰 OpenAI Agent & Tools

Uses:

• search_docs: Queries knowledge base
• add_docs: Uploads new documents
• update_doc: Updates vectorized entries
• delete_doc: Removes documents from the vector DB

4. 🔗 Multimodal Ingestion Pipeline

Handles file ingestion and embedding:

• Text extraction (pytesseract, python-docx)
• Embedding with text-embedding-3-small
• Insertion into Qdrant
• Result logging to JSON

5. 🧠 Databases

• Qdrant Cloud: Vector database for storing embeddings
• MongoDB Atlas: Stores tickets and user interactions

---

🚀 Getting Started

1. 📥 Clone the Repo

git clone https://github.com/anivenk25/Cervello.git
cd Cervello_server

2. 📦 Install Dependencies

We use uv for virtual env management.

Ensure Python >= 3.10 is installed.

• For macOs/Linux

curl -LsSf https://astral.sh/uv/install.sh | sh

• For windows

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

• Using pip

pip install uv

3. 🔐 Setup Environment Variables

Create a .env file in the root and configure: (refer to env.local

OPENAI_API_KEY=your_openai_key
QDRANT_API_KEY=your_qdrant_key
QDRANT_URL=https://your-qdrant-url
MONGO_URI=mongodb+srv://...

4. 🔐 Activate virtual env

• For windows

.venv\Scripts\activate

• For Linux

source .venv/bin/activate

5. Install dependencies

uv pip install -r requirements.txt

5. Run the venv

1. To run the server for queries and db updation/deletion

uvicorn mcp_server:app --reload --log-level debug

1. To run the server for ticketing

uvicorn tickets:app --reload --log-level debug

---

📂 Directory Structure

.
├── main_app/
│   ├── app.py                 # FastAPI RAG endpoints
│   ├── ingestion/             # File parsing and embedding
│   └── agents/                # OpenAI agent tools
|   └── tickets.py
|   └── mcp_server.py 
├── tickets/
│   └── tickets.py             # Ticketing FastAPI service
├── examples/
│   └── gpt_4o_multimodal_rag/ # Demo apps and workflows
├── .env                       # Environment variables
├── pyproject.toml             # Poetry config
└── README.md

---

🧪 How to Use

➕ Upload Documents

• Send .pdf, .docx, or .png files to the /upload endpoint.
• Text will be extracted and embedded via OpenAI.
• Stored in Qdrant for semantic retrieval.

🔍 Search with RAG

• Hit /query with a natural language question.
• Top-k relevant documents retrieved from Qdrant.
• OpenAI responds with a generated answer.

🧠 Agent Mode

• Use /mcp/query for enhanced query with tool calling.
• OpenAI auto-selects appropriate tool like upload/search/update.

🎫 Create Tickets

• The system may suggest creating a ticket.
• User can edit and submit via /create-ticket.
• Stored in MongoDB and retrievable anytime.

---

📌 Requirements

• Python ≥ 3.10
• Poetry
• MongoDB Atlas account
• Qdrant Cloud account
• OpenAI API Key

---

🛠️ Tech Stack

Component	Technology
API Backend	FastAPI
Vector Store	Qdrant
Embeddings	OpenAI text-embedding-3-small
Agent LLM	OpenAI GPT-3.5/GPT-4o
Database	MongoDB Atlas
Text Parsing	Pathway, pytesseract, python-docx
Frontend Client	(Optional) Chat interface or Postman

---

👥 Contributing

We welcome PRs, feedback, and issues. Please:

1. Fork this repo
2. Create a new branch
3. Push your changes
4. Open a Pull Request

---

📄 License

MIT License

---

📫 Contact

Feel free to reach out with questions or collaboration requests:

• Anirudh Venkateswaran
• Srishty M
• Rishi Das
• Organization: Cerevello