CardCapture LLC runs a platform for college fairs, where recruiters used to type up stacks of handwritten student inquiry cards by hand. I built the AI processing pipeline on FastAPI in three stages: extraction, enhancement, and review. Google Document AI handles OCR, and Gemini cleans and structures the fields. Six sequential enhancers then map fields, split names, validate addresses through Google Maps, and match high schools to their CEEB codes. I added a duplicate check before the expensive Gemini call to keep AI costs down. The result was 95% field accuracy and processing about 10x faster than manual entry.