Ask questions of your own documents with a local LLM. Nothing leaves the machine.

role
Solo build
status
Shipped
when
2025
ownership
Solo
Repository
rag.qaAI research · Intelligent systems
.pdf.docx.mdFAISS · top-k

Who did what. Public repository; the approach matches the RAG QA system I built at ZSapiens.

Overview

A retrieval-augmented chatbot that ingests PDF, DOCX, TXT and Markdown into a local FAISS index and answers with GPT4All, served by FastAPI with an optional Gradio front end.

Why it exists

Teams want answers from their own documents without sending them to a hosted model.

My part

  • Multi-format ingestion and chunking into a FAISS vector store with SentenceTransformers embeddings.
  • Retrieval and local generation with GPT4All behind a FastAPI service.

How it works

  1. a

    Ingest

    PDF, DOCX, TXT and MD into chunks.

  2. b

    Embed

    SentenceTransformers (all-MiniLM-L6-v2).

  3. c

    Retrieve

    FAISS nearest neighbours.

  4. d

    Answer

    GPT4All generates from retrieved context.

Built with

  • FastAPI
  • FAISS
  • SentenceTransformers
  • GPT4All
  • Gradio

Status & next

Runs fully offline.

Source: github.com/AA-KH/RAG-Chatbot and résumé (ZSapiens)