Librarium

A local RAG-based Document Q&A system

Chat with your documents. Runs fully offline. Zero API costs.

---

Overview

Librarium is a fully offline document Q&A system built on a RAG (Retrieval-Augmented Generation) pipeline. Upload PDFs, Word docs, text files, or paste any URL — then ask questions in plain English and get accurate, cited answers powered by a local LLM.

No API keys. No cloud dependency. No cost.

---

Preview

---

Features

Feature	Description
Hybrid Search	Combines BM25 keyword search and vector search for better retrieval
Multi-format Support	PDF, DOCX, TXT, and web URLs
OCR	Extracts text from image-based and scanned PDFs automatically
Conversation Memory	Ask follow-up questions — the system remembers context
Citations	Every answer shows the exact source file and page number
Query Rewriting	Vague questions are automatically rewritten for better retrieval
Auto Summary	Generates a document summary on load
URL Support	Fetch and index any webpage mid-session
Chat Export	Export your Q&A session as a formatted PDF
Eval Dashboard	Live grounding score, response time, and retrieval stats per query
Streaming	Answers stream word-by-word for a responsive feel
Persistent Storage	Vector DB saves to disk and reloads instantly on future runs

---

Tech Stack

Layer	Technology
Orchestration	LangChain
LLM	Ollama (llama3.2, llama3.2:1b) — runs locally
Embeddings	Ollama (nomic-embed-text)
Vector DB	ChromaDB
Keyword Search	BM25 (rank-bm25)
OCR	Tesseract + pdf2image
Web Scraping	BeautifulSoup + Requests
UI	Streamlit

---

Setup

Prerequisites

• Python 3.10+
• Ollama installed and running
• Tesseract OCR (Windows installer)
• Poppler (Windows)

Installation

1. Clone the repository

git clone https://github.com/keya115251/document-qna.git
cd document-qna

2. Create and activate a virtual environment

python -m venv venv
venv\Scripts\activate       # Windows
source venv/bin/activate    # Mac/Linux

3. Install dependencies

pip install -r requirements.txt

4. Pull the Ollama models

ollama pull llama3.2
ollama pull llama3.2:1b
ollama pull nomic-embed-text

5. Update paths in app.py

TESSERACT_PATH = r"C:\Program Files\Tesseract-OCR\tesseract.exe"
POPPLER_PATH   = r"C:\poppler\poppler-26.02.0\Library\bin"

6. Run

streamlit run app.py

---

Usage

• Upload a PDF, DOCX, or TXT file via the sidebar
• Or paste any URL to index a webpage
• Ask questions in the chat input
• Type load url <url> in chat to add more sources mid-session
• Toggle the Eval Dashboard to see retrieval metrics
• Export your session as a PDF from the sidebar

---

Roadmap

• Hybrid search (BM25 + vector)
• Multi-document support
• OCR for image-based PDFs
• Query rewriting
• Citations with source and page attribution
• Conversation memory
• URL support with mid-session loading
• Auto document summary
• Chat history export as PDF
• Streamlit Web UI
• Evaluation dashboard

---

_{Built by Keya Chembuli}