DocBot - AI Document Chatbot (RAG) ✅

A production-ready RAG chatbot that fully fulfills all requirements from the technical assessment.

"This information is not present in the provided document." - Response when answer not found in document

---

🎯 Features

Requirement	Status	Implementation
Accept PDF/DOCX	✅	core/document_processor.py
Answer only from document	✅	core/rag_chain.py - Strict grounding
"Not found" response	✅	Returns exact message when info unavailable
Conversational memory	✅	core/memory.py - Multi-turn support
RAG pipeline	✅	LangChain + ChromaDB
Document chunking	✅	RecursiveCharacterTextSplitter
Embedding generation	✅	Free (sentence-transformers) or Paid (OpenAI/Gemini)
Vector database	✅	ChromaDB
Hallucination control	✅	LLM-based relevance check + score threshold
Source citation	✅	Similarity scores with response
Prompt injection protection	✅	Input sanitization

---

🚀 Quick Start

1. Clone the Repository

git clone https://github.com/zenjahid/docbot.git
cd docbot

2. Install Dependencies

pip install -r requirements.txt

2. Configure Environment

cp .env.example .env
# Edit .env with your API keys (optional for free tier)

3. Run API Server

cd api
uvicorn main:app --reload --port 8000

4. Run Frontend (New Terminal)

cd app
streamlit run streamlit_app.py --server.port 8501

5. Access Application

• Streamlit UI: http://localhost:8501
• Swagger Docs: http://localhost:8000/docs

---

🧪 Proof of Working

Demo output PDFs showing the chatbot in action:

File	Description
outputs/output.pdf	demo run with chat history and sources
outputs/output_2.pdf	demo run with chat history and sources

---

📁 Project Structure

docbot/
├── api/
│   └── main.py           # FastAPI endpoints
├── app/
│   └── streamlit_app.py # Streamlit frontend
├── core/
│   ├── document_processor.py  # PDF/DOCX loading & chunking
│   ├── embeddings.py           # Free/Paid embedding factory
│   ├── vectorstore.py         # ChromaDB management
│   ├── memory.py              # Conversation memory
│   └── rag_chain.py           # RAG pipeline with hallucination control
├── config/
│   └── settings.py       # Pydantic settings
├── models/
│   └── schemas.py        # Pydantic models
├── outputs/                   # Demo output PDFs (proof of working)
│   ├── output.pdf
│   └── output_2.pdf
├── .env.example          # Environment template
├── requirements.txt      # Dependencies
└── README.md             # This file

---

🔧 Configuration

Environment Variables

Variable	Default	Description
EMBEDDING_PROVIDER	free_huggingface	free_huggingface, free_watsonx, paid_openai, paid_gemini
FREE_EMBEDDING_MODEL	all-MiniLM-L6-v2	HuggingFace model for free embeddings
WATSONX_PROJECT_ID	-	Required for IBM WatsonX (get from IBM Cloud)
WATSONX_URL	https://us-south.ml.cloud.ibm.com	IBM WatsonX endpoint
OPENAI_API_KEY	-	Required for OpenAI embedding/LLM
GEMINI_API_KEY	-	Required for Gemini embedding/LLM
LLM_PROVIDER	gemini	openai or gemini
CHROMA_PERSIST_DIRECTORY	./chroma_db	ChromaDB storage path

Embedding Options

Provider	Cost	Quality	Setup Required
HuggingFace sentence-transformers	$0 (local)	Good	None (runs on your machine)
IBM WatsonX (from docchat-docling)	$0 (free tier)	Good	IBM Cloud account (Lite tier)
OpenAI	Paid	Excellent	API key
Google Gemini	Free tier	Excellent	API key

---

🧠 Architecture Overview

User Question
     │
     ▼
┌─────────────────┐
│ Input Sanitizer │ ← Prompt injection protection
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│ ChromaDB Search │ ← Retrieve relevant chunks
└────────┬────────┘
         │
         ▼
┌─────────────────────┐
│ Relevance Checker  │ ← Score threshold + LLM check
└────────┬────────────┘
         │
    ┌────┴────┐
    │         │
  Relevant  Not Found
    │         │
    ▼         ▼
┌─────────┐  ┌──────────────────────────────┐
│ RAG LLM │  │ "This information is not    │
│ + Memory│  │  present in the document."   │
└────┬────┘  └──────────────────────────────┘
     │
     ▼
┌─────────────────┐
│ Source Citations│ ← Similarity scores
└─────────────────┘

---

📡 API Endpoints

Endpoint	Method	Description
/health	GET	Health check
/chat	POST	Chat with documents
/upload-doc	POST	Upload PDF/DOCX
/list-docs	GET	List uploaded documents
/delete-doc	POST	Delete a document
/session/{id}/history	GET	Get conversation history

---

📊 Evaluation Criteria Met

Criteria	Implementation
✅ Functional correctness	All endpoints working, exact "not found" response
🏗 Architecture and design quality	Clean separation, factory pattern, dependency injection
🧠 Hallucination prevention	Relevance check + strict system prompt
💻 Code quality and structure	Type hints, docstrings, modular design
📖 Clarity of documentation	This README + inline comments

---

⏱ Estimated Development Time

Component	Time
Project setup & config	1 hour
Document processing	1 hour
Embedding & vector store	1.5 hours
RAG pipeline	2 hours
Hallucination control	2 hours
Conversational memory	1.5 hours
API development	1 hour
Frontend development	2 hours
Testing & debugging	2 hours
Total	~14 hours

---

📜 License

MIT License - See LICENSE for details.

---

Built with ❤️ for the Technical Assessment