Clarifile

Clarifile is a lightweight Retrieval-Augmented Generation (RAG) system built with ChromaDB, SentenceTransformers, and an LLM (local Ollama Llama 3.1 or a hosted provider). It retrieves relevant chunks from a prebuilt vector store and generates grounded answers with citations only when sources are used. A minimal Streamlit app is included.

Features

• PDF & URL ingestion with text cleaning and chunking
• Persistent vector storage via ChromaDB
• Top-k semantic search with scores
• Context construction and deduplication
• LLM query with local 'llama3.1:8b'
• Automatic source citations
• Minimal Streamlit UI: KB summary + question box

Structure

• ingest.py # Ingest PDFs/URLs
• retriever.py # Retrieval + QA flow
• data/
  • (pdf files)
• storage/
  • (collection data)

Requires

• Python ≥ 3.10
• SentenceTransformers all-MiniLM-L6-v2
• ChromaDB (persistent client)
• One LLM path:
  • Local: Ollama with llama3.1:8b
  • Hosted: Groq API (e.g., llama-3.1-8b-instant)

Prebuild the Knowledge Base (Required)

The app expects a ready vector store in ./storage/. Preload all PDFs in data/Demo/ and URLs from data/Demo/urls.txt:

Configure the LLM

Clarifile routes all generations through retriever.answer() → call_llm().

• Hosted (Streamlit Cloud / public demo):
  • Set secrets/env:
  • GROQ_API_KEY — Groq API key
  • MODEL_ID — e.g., llama-3.1-8b-instant
• Local (developer laptop):
  • Install Ollama and pull the model:
  • ollama pull llama3.1:8b

Leave GROQ_API_KEY unset to use http://localhost:11434.

Installation

'''bash pip install -r requirements.txt

Usage

'''python from ingest import ingest_pdfs, ingest_urls from retriever import answer

ingest_pdfs(["data/testPDF1p.pdf", "data/testPDF3p.pdf"]) ingest_urls([ "https://www.example.com", "https://jsonplaceholder.typicode.com/todos/1" ])

print(answer("What does the example site contain?"))

Run the App

'''bash streamlit run app.py

• The app lists detected demo PDFs and URLs.
• It requires ./storage/ to exist; if missing, build it first (see Prebuild).
• Ask a question; answers cite sources only when relevant chunks were used. or
• Scan the qr code to test app

Notices

Built with Llama 3.1

Llama 3.1 is licensed under the Llama 3.1 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved.
See the Llama 3.1 Community License and Acceptable Use Policy for more information.