🔍 Intelligent Forensic Document Auditor (Local GenAI)

An automated data integrity tool designed to identify discrepancies and unauthorized alterations in high-stakes corporate documents. This project was developed as a Proof of Concept (POC) for the Digital Accelerator division to ensure 100% data privacy and eliminate cloud-based API costs.

🎯 Target Use Case

• Forensics: Identifying tampered text, modified dates, or altered financial figures in official records.
• Legal/Finance: Automating the review of complex, multi-page contracts (10+ paragraphs) for data consistency.
• Process Automation: Reducing manual auditing time by 90% while maintaining enterprise-grade data security.

🛠️ Tech Stack

• Language: Python 3.11+
• LLM Orchestration: LangChain (LCEL)
• Local LLM: Llama 3.2 (via Ollama)
• Embeddings: nomic-embed-text (Local via Ollama)
• Vector Database: ChromaDB (with MMR search logic)
• Frontend: Streamlit for data visualization and interaction

🚀 Key Technical Solutions

• Privacy-First Architecture: By using Ollama, the system processes documents entirely offline, making it suitable for sensitive internal communications.
• Advanced Retrieval: Implemented Maximum Marginal Relevance (MMR) and adjusted retrieval parameters ($k=6$) to ensure accuracy in long-form documents.
• Stability Engineering: Resolved Windows-specific environment conflicts, including telemetry bugs and file-locking errors, through dynamic session pathing.

📋 Setup & Installation

1. Install Ollama: Download from ollama.com and pull the required models:

   ollama pull llama3.2
   ollama pull nomic-embed-text

2. Environment Setup:

   python -m venv audit_env
   # Windows
   .\audit_env\Scripts\activate
   # Install dependencies
   pip install -r requirements.txt

3. Run Application:

   streamlit run app.py

---

Developer: Sethukb
Portfolio: codes-by-sethu.github.io/PORTFOLIO/