AI Extraction Evaluator and Data Labeling Analyst - Bank Statement Parsing
Used Google Gemini Vision AI to extract structured data from multi-page bank statement PDFs, requiring per-page accuracy validation and correction. Implemented a math validation loop to cross-check totals and auto-retry extraction on error, ensuring high data accuracy for AI model output. Benchmarked alternative document vision models to evaluate and compare extraction quality and model performance. • Labeled bank statement documents by evaluating page-by-page extraction accuracy. • Performed math validation on extracted financial data to ensure high-quality labels. • Compared multiple vision AI providers and documented extraction flaws for improvement. • Managed outputs in structured formats (Excel/CSV/JSON) for further model retraining or evaluation.