🔍 Audit RAG System (LangChain + Ollama)

A local Retrieval-Augmented Generation (RAG) system built for audit and transaction analysis.
It allows users to query official audit manuals and receive accurate, context-grounded answers generated strictly from source documents.

Designed as a secure, offline, enterprise-ready RAG proof-of-concept.

---

✨ Highlights

• 📄 Queries real audit manuals (PDFs)
• 🧠 Uses Phi LLM for fast local inference
• 🔎 Semantic search via FAISS + Nomic embeddings
• 🚫 Hallucination-safe (answers only from retrieved context)
• 💻 Fully local — no cloud APIs

---

🧠 Tech Stack

Component	Choice
LLM	phi (Ollama)
Embeddings	nomic-embed-text
Vector DB	FAISS
Framework	LangChain
Language	Python
Runtime	Ollama (local)

---

📚 Knowledge Base

The system is trained on official audit and transaction analysis guides, including:

• Split Transactions
• Incorrect Totals
• Duplicate Transactions
• Suspicious Date Ranges
• Transaction Aging
• Fuzzy Name Matching
• Statistical Anomaly Detection

📦 11 audit manuals (~5.2 MB) stored locally and indexed into a FAISS vector store.

---

🏗️ Project Structure

RAG_for_Audit/
├── Manuals/          # Audit PDF documents
├── Audit_FIASS/      # Persisted FAISS vector store
├── RAG Model.ipynb   # Indexing + retrieval pipeline
├── bot.py            # CLI-based audit assistant
└── README.md