DocLens 🔍

Ask questions across all your entire document library. Run compliance audits. Get answers with source citations.

---

Overview

DocLens turns your document library into a searchable knowledge base. Upload PDFs, ask questions across all files, and run automated compliance checks. Works locally for privacy or in the cloud for speed.

• Built for teams dealing with contracts, policies, compliance docs, and internal knowledge bases.

📸 Interface

🔒 Private Mode (Local) Secure processing using local LLM.	☁️ Cloud Mode (Performance) High-speed processing using Gemini.
Audit Configuration Select target documents and rule sets.	Audit Report Clear report showing violations and evidence.

🔻 View Document Library

Manage your uploaded files and rule sets.

---

Key Features

🧩 Advanced Hybrid Search

• Dual-Stage Retrieval: Combines Semantic Search (Dense Vectors) with Keyword Search (BM25) for maximum accuracy.
• Reciprocal Rank Fusion (RRF): Re-ranks results to ensure the most relevant context is prioritized.

🛡️ Switch Between Local & Cloud AI

• 🔒 Private Mode (Ollama): Runs entirely offline. Best for sensitive financial or legal documents.
• ☁️ Cloud Mode (Gemini): High-performance reasoning for general knowledge tasks.

🧾 Automated Compliance Auditor

• Rule-Based Checking: Compare any target document against a defined "Rule Set" (e.g., Company Policy).
• Structured Reporting: Generates readable cards with evidence citations.

🧠 Shared Knowledge Base

• Centralized Library: Upload, store, and manage metadata for all business PDF documents.
• Categorization: Classify files as "General Knowledge" or "Rule Sets" to guide the retrieval logic.

💬 Cross-Document Q&A

• Library-Wide Search: Ask questions and get answers from across your entire document library, not just single files.
• Citations: Every answer includes links to the source text so you can verify the facts.
• Real-Time Streaming: Answers appear instantly as they are generated, rather than waiting for the whole block of text.

---

Technical Stack

Component	Technology	Role
Orchestration	LangChain	Chain logic, Document Loaders, Splitters
Vector DB	Qdrant (Docker)	Storing embeddings & metadata
Retrieval	BM25 + RRF	Hybrid Search Algorithms
Inference	Ollama / Gemini	Local & Cloud LLM providers

---

How to Run

1. Clone the repository:

   git clone https://github.com/sdave0/doclens.git

2. Start the Vector Database:

   docker-compose up -d

3. Install Dependencies:

   pip install -r requirements.txt

4. Set Environment Variables: Create a .env file:

   GOOGLE_API_KEY=your_key_here
   QDRANT_URL=http://localhost:6333

5. Launch App:

   streamlit run app.py

---