Advance RAG Studio

A modern Retrieval-Augmented Generation (RAG) app built from your notebook workflow.

What You Get

• main.py: reusable backend engine for ingestion, chunking, indexing, retrieval, and answer generation
• app.py: modern Streamlit UI with polished styling, chat layout, and live system status cards
• requirements.txt: project dependencies

Features

• Ingest files: .txt, .md, .pdf, .wav, .mp3, .mp4, .m4a, .flac, .ogg, .png, .jpg, .jpeg, .webp
• Audio transcription via Whisper
• Text chunking by words
• Multimodal RAG with image captions (Groq vision model)
• Embeddings via Jina Embeddings v4 API
• Vector search with FAISS
• Cross-encoder reranking for improved relevance
• Guardrails for context-only answers
• Session memory and latency tracking in the UI
• Optional vision-at-answer for image Q&A plus image previews in context
• Metadata filtering (text-only / image-only / both)
• Dashboard-style UI with status cards and tabbed workflows
• Quick prompts, adjustable answer temperature, and downloadable chat history (JSON)
• Source Library tab to inspect indexed chunks by source/modality
• Retrieval Inspector tab to test candidate retrieval and reranked results
• Source-level analytics (chunks/text/image counts per file)
• Remove selected sources and auto-rebuild index
• Manual index rebuild action from the sidebar

Setup

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

Set your API keys:

set GROQ_API_KEY=your_groq_api_key_here
set JINA_API_KEY=your_jina_api_key_here

Or provide them in the Streamlit sidebar at runtime.

Run UI

streamlit run app.py

Project Structure

Advance_RAG/
  app.py
  main.py
  requirements.txt
  README.md
  Advance_RAG(11_02_2026).ipynb

Backend Usage (Optional)

from main import AdvanceRAG

rag = AdvanceRAG(groq_api_key="your_key")
rag.ingest_txt_bytes(open("notes.txt", "rb").read(), source="notes.txt")
rag.build_index()
print(rag.answer("Summarize the key topics"))

Notes

• Do not hardcode API keys in source files.
• First run of Whisper may download model weights, so audio transcription can take longer.
• For large files, prefer smaller top_k and tune chunk_size from the UI sidebar.
• Cross-encoder reranking uses sentence-transformers and can add startup latency on first load.

Streamlit Cloud Deployment

1. Push this repo to GitHub.
2. In Streamlit Cloud, set app.py as the entrypoint.
3. Add secret in app settings:

GROQ_API_KEY = "your_groq_api_key_here"
JINA_API_KEY = "your_jina_api_key_here"

4. Deploy.

Notes:

• runtime.txt pins Python for better binary wheel compatibility on Streamlit Cloud.
• Audio transcription needs both openai-whisper and ffmpeg; if unavailable, the app still runs and skips audio files.