Skip to content

Latest commit

 

History

2 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 

Repository files navigation

QUANDOC AI

Enterprise-Grade Document Intelligence Platform powered by RAG, Multi-Agent Orchestration, and Local LLMs.


Overview

QUANDOC AI is an advanced AI-powered document intelligence system designed to understand, retrieve, analyze, compare, and reason over uploaded documents using Retrieval-Augmented Generation (RAG).

The system combines semantic retrieval, hybrid search pipelines, reranking strategies, orchestration workflows, memory-aware conversations, and local LLM inference to create a production-style enterprise AI architecture.

Unlike traditional chatbot demos, QUANDOC AI focuses on:

  • grounded AI responses
  • hallucination reduction
  • retrieval quality
  • intelligent orchestration
  • enterprise-style reasoning workflows

Core Features

  • Advanced RAG Architecture
  • Hybrid Retrieval (Semantic + BM25)
  • Pinecone Vector Database Integration
  • Multi-Agent Workflow Orchestration
  • LangGraph AI Flows
  • Query Rewriting Pipeline
  • Context-Aware Memory System
  • Reranking Pipeline
  • Hallucination Control
  • Local LLM Inference using Ollama
  • PDF Document Intelligence
  • Conversational Document QA
  • Enterprise-Style AI Reasoning

Tech Stack

Backend

  • FastAPI
  • Python
  • Uvicorn

AI / LLM Stack

  • LangChain
  • LangGraph
  • Ollama
  • Phi3 / Llama3 / Mistral
  • Nomic Embeddings

Retrieval & Vector Search

  • Pinecone
  • Hybrid Retrieval
  • BM25
  • Semantic Search
  • Reranking Pipelines

AI Engineering Concepts Implemented

  • Retrieval-Augmented Generation (RAG)
  • Multi-Agent Systems
  • AI Orchestration
  • Query Rewriting
  • Context Grounding
  • Hallucination Reduction
  • Memory-Aware Conversations
  • Workflow Routing
  • Intelligent Context Synthesis

System Architecture

User Query
   ↓
Query Rewriter
   ↓
Hybrid Retriever
(BM25 + Semantic Search)
   ↓
Reranker
   ↓
Context Builder
   ↓
LangGraph Orchestrator
   ↓
AI Agent Workflow
   ↓
LLM Response Generation
   ↓
Grounded Enterprise Response

Project Structure

backend/
│
├── app/
│   ├── api/
│   ├── rag/
│   ├── workflows/
│   ├── agents/
│   ├── services/
│   └── main.py
│
├── uploads/
├── requirements.txt
└── .env

Installation

1. Clone Repository

git clone https://github.com/yourusername/quandoc-ai.git

cd quandoc-ai/backend

2. Create Virtual Environment

python -m venv venv

Activate Environment

Windows

venv\Scripts\activate

Mac/Linux

source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

Environment Variables

Create a .env file:

PINECONE_API_KEY=your_api_key
PINECONE_INDEX=quandoc-ai

Install Ollama

Download:

https://ollama.com

Pull models:

ollama pull phi3

ollama pull nomic-embed-text

Run Backend

uvicorn app.main:app --reload

Server:

http://127.0.0.1:8000

Swagger Docs:

http://127.0.0.1:8000/docs

Testing QUANDOC AI

Upload PDF

Use:

  • /upload
  • Upload research paper or enterprise document

Test RAG Query

Example JSON:

{
  "question": "Explain the hybrid retrieval architecture discussed in the document."
}

Advanced Testing Questions

Retrieval Quality

{
  "question": "Summarize the main architecture discussed across multiple sections."
}

Memory Testing

{
  "question": "Compare this section with the previous concept discussed earlier."
}

Hallucination Testing

{
  "question": "What does the document say about quantum blockchain insurance?"
}

Expected:

  • system should refuse unsupported answers

AI Engineering Challenges Solved

While building QUANDOC AI, several real-world AI engineering problems were addressed:

  • hallucinated responses
  • retrieval inconsistency
  • vector database integration
  • reranking quality
  • workflow orchestration failures
  • context fragmentation
  • dependency conflicts
  • memory handling
  • retrieval latency optimization

Key Learnings

QUANDOC AI was built as a hands-on AI engineering system to deeply understand:

  • production-grade RAG systems
  • enterprise AI orchestration
  • retrieval pipelines
  • LLM reliability
  • system-level AI debugging
  • multi-agent workflow design

The project reinforced an important lesson:

Real AI engineering is learned by building systems that break — and solving the problems behind them.


Future Improvements

  • Streaming responses
  • Citation generation
  • Multi-modal document support
  • OCR pipelines
  • Advanced agent routing
  • Redis memory layer
  • Kubernetes deployment
  • Authentication & RBAC
  • Real-time collaboration
  • Evaluation & observability dashboards

Author

Uzain Mohid

AI Engineering • RAG Systems • GenAI Workflows • Enterprise AI Architecture


About

Enterprise-grade AI document intelligence platform powered by RAG, LangGraph, Pinecone, hybrid retrieval, and local LLM orchestration.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages