johnsonhk88/AI-Bank-Statement-Document-Automation-By-LLM-And-Personal-Finanical-Analysis-Prediction

AI Bank Statement Document Automation By LLM model and Personal Finanical Analysis

What it solves

This project automates the extraction and analysis of financial data from bank statements. It solves the challenge of converting unstructured PDF documents into structured financial insights, while ensuring privacy by redacting personally identifiable information (PII) before data is stored.

How it works

The system uses a multi-stage pipeline combining computer vision and language models:

  1. Document Parsing: It employs YOLO for layout detection and OCR to extract text from PDFs.
  2. Agentic Workflow: CrewAI agents with specialized skills handle parsing, PII redaction, and financial analysis.
  3. Local LLM Integration: It uses LiteLLM to connect to local models (via LM Studio or Ollama) to maintain data privacy and reduce costs.
  4. RAG Pipeline: Redacted data is embedded into vector databases like Qdrant or Chroma, allowing users to query their financial history using natural language.
  5. Monitoring: MLflow is used to trace and debug the agent workflows and LLM calls.

Who it’s for

  • Individuals wanting to automate their personal finance tracking.
  • Developers looking for a reference implementation of a local-first RAG pipeline for sensitive documents.
  • Users who prefer running LLMs locally to keep financial data private.

Highlights

  • Privacy-First: Redacts PII before embedding data into the vector database.
  • Local-First: Optimized for local LLMs (9B+ parameters) via LM Studio and Ollama.
  • Local LLM Guidance: Provides specific recommendations on model size and context length for optimal performance.
  • Comprehensive Tooling: Integrates YOLO, OCR, CrewAI, and MLflow for a professional-grade automation pipeline.