AI-Powered Handwritten Prescription Recognition System (Award-Winning)

● Engineered a Healthcare AI system that recognizes handwritten and printed medical prescriptions and extracts medicine names using a fine-tuned TrOCR (Transformer-based OCR) model, deployed via FastAPI with GPU acceleration and automatic CPU fallback.

● Designed a fuzzy-matching NLP pipeline (RapidFuzz) that maps noisy OCR output to standardized RxNorm generic and brand drug names, returning RXCUI codes, term-type (TTY) classification, and confidence scores.

● Built an end-to-end data pipeline that parses and cleans raw RxNorm RRF/CSV records, including deduplication, generic/brand tagging, and ingredient-brand mapping, into a production-ready drug database.

● Deployed documented REST API endpoints (/predict for image upload, /match-text for direct text lookup); the project won 3rd Place in a university-wide graduation project competition.