KNOWLEDGE BASE AI AGENT (LOCAL RAG SYSTEM)

An offline AI Knowledge Base Agent that answers questions from PDF, TXT, and DOCX documents using real Retrieval-Augmented Generation (RAG). Built in 48 hours using FAISS vector search, semantic chunking, and Mistral-7B-Instruct, with a clean Streamlit UI. Runs fully locally on an RTX 3050 (6GB VRAM) — no API keys required.

FEATURES

• Upload multiple documents at once (PDF, TXT, DOCX)
• Real RAG pipeline using FAISS vector DB + SentenceTransformers
• Mistral-7B-Instruct for local inference
• Semantic chunking for accurate retrieval
• Structured responses with: [SUMMARY], [DETAILED EXPLANATION], [CONFIDENCE]
• Shows source and page numbers
• Completely offline & privacy-friendly

TECH STACK

UI: Streamlit Embedding Model: SentenceTransformer (all-MiniLM-L6-v2) LLM: Mistral-7B-Instruct Vector Store: FAISS File Support: PDF, TXT, DOCX PDF Processing: PyPDF2 + Semantic Chunking DOCX Processing: python-docx Backend Language: Python

SETUP INSTRUCTIONS

git clone https://github.com/YOUR_USERNAME/knowledge-base-agent.git cd knowledge-base-agent pip install -r requirements.txt streamlit run knowledge_base_agent/app.py

ARCHITECTURE OVERVIEW

Upload Documents → Chunk & Embed → FAISS Search → Mistral RAG → Structured Answer

FUTURE IMPROVEMENTS

• CSV, XLSX, PPTX support

• Conversational chat history

• Document-wide summary mode

• HR agent / Support agent extensions

• API deployment support

    ┌───────────────────────────┐
    │       Upload Files        │
    │   (PDF / TXT / DOCX)      │
    └─────────────┬─────────────┘
                  │
                  ▼
    ┌───────────────────────────┐
    │   Text Extraction Layer   │
    │ PyPDF2 / python-docx / IO │
    └─────────────┬─────────────┘
                  │
                  ▼
    ┌───────────────────────────┐
    │  Semantic Chunking        │
    │ RecursiveCharacterSplitter│
    └─────────────┬─────────────┘
                  │
                  ▼
    ┌───────────────────────────┐
    │ Create Embeddings         │
    │ SentenceTransformer Model │
    └─────────────┬─────────────┘
                  │
                  ▼
    ┌───────────────────────────┐
    │ FAISS Vector Search (RAG) │
    └─────────────┬─────────────┘
                  │
                  ▼
    ┌───────────────────────────┐
    │   Mistral-7B-Instruct     │
    │  Structured Answer Output │
    └─────────────┬─────────────┘
                  │
                  ▼
    ┌───────────────────────────┐
    │ Streamlit UI              │
    │ Display Final Answer      │
    └───────────────────────────┘
  

FINAL NOTE

This project demonstrates real RAG implementation, modular document ingestion, and local AI deployment potential. Focused on practical agent development rather than UI decoration.