KnowBridge — RAG-Powered Knowledge Assistant

KnowBridge is a lightweight Retrieval-Augmented Generation (RAG) app that turns a set of documents into a grounded, searchable knowledge assistant.

It provides:

• A Gradio Web UI to upload and index .md, .docx, and .doc files
• Persistent ChromaDB vector storage (no re-embedding every run)
• Semantic retrieval + grounded answering using a Groq-hosted LLM
• Session-based chat persistence (SQLite) with rolling summarization for long chats

Project scope

KnowBridge is designed to work with user-supplied document collections across domains rather than being tied to a single business function.

Supported usage:

• Upload .md, .docx, or .doc files
• Ask factual, procedural, or follow-up questions answerable from those documents
• Receive grounded answers with a SOURCES: line for traceability

Out of scope:

• Questions requiring external knowledge not present in the uploaded documents
• Live data lookups from external systems
• Formal benchmark-grade retrieval evaluation in the current version

How it works (brief)

1. Upload .md, .docx, or .doc files in the Knowledge Base tab.
2. The app hashes each file and only re-indexes files that changed.
3. Documents are chunked, embedded with a sentence-transformer model, and stored in ChromaDB.
4. In the Chat tab, your question is embedded, relevant chunks are retrieved, and the LLM answers using only retrieved context.
5. The assistant appends a SOURCES: line for traceability and stores chat history in SQLite.

Requirements

Software

• Python 3.11 recommended

Note: Some dependencies (ChromaDB / PyTorch / sentence-transformers) may not have wheels for very new Python versions yet (e.g. 3.14). If your python3 is too new, use python3.11.

Python libraries

Install all dependencies from this folder’s requirements file:

pip install -r requirements.txt

Key packages include: langchain, langchain_groq, chromadb, sentence-transformers, gradio, python-dotenv, pyyaml, python-docx.

API keys

This app requires a Groq API key.

Create a .env file in the project root (this folder) with:

GROQ_API_KEY=your-groq-api-key

Get a key from: https://console.groq.com/

How to run

From the rag-knowledge-assistant/ directory:

python3.11 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

python code/app.py

Then open the local Gradio URL printed in the terminal (typically http://127.0.0.1:7860).

Configuration

• App settings (LLM model, retrieval params): code/config/config.yaml
• Prompt rules (grounding, SOURCES: formatting): code/config/prompt_config.yaml

Common knobs:

• vectordb.n_results (top-k retrieval)
• vectordb.threshold (distance threshold)
• llm (Groq model name)

Data and persistence

• Sample knowledge docs: data/ (sample .md files included)
• Vector DB persisted to: outputs/vector_db/
• Chat history persisted to: outputs/chat_history.db

Notes

• The UI supports indexing .md, .docx, and .doc.
• For .doc, the app will try to extract text from Confluence/Atlassian export formats (MIME-wrapped HTML) first, then fall back to macOS textutil or LibreOffice (soffice) conversion.
• If retrieval returns no chunks under the threshold, the pipeline falls back to returning the top-k results to avoid empty context.

License & attribution

This project is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).

• See LICENSE for the full text.
• Attribution: If you share, reuse, or adapt this project, please provide attribution.
  • Credit: Raja (this KnowBridge implementation) and retain the original license notice.