Jin Khye Tan

Design, develop, and evaluate OCR techniques for financial documents and table extraction across 15+ banks, supporting SME loan analytics workflows. Implement and benchmark OCR models, vision-language models, and PDF parsers, hosting open-source systems using vLLM and LLaMA-CPP for performance comparison. Integrate key information and table extraction modules into the production pipeline, conducting testing and documentation to ensure reliability and scalability. Led prompt engineering efforts and evaluated schema consistency across 100+ annual report runs to ensure structured and reliable outputs. Developed modular Gen AI backend pipelines using FastAPI, LangChain, vLLM, and YAML prompts to automate financial data extraction from handwritten notes. Led production deployment and benchmarking with vLLM.

Containerized existing demos using Docker and Docker Compose for quick environment setup and teardown. Authored 70+ pages of technical documentation and led knowledge transfer sessions, enabling seamless handover and maintainability.

Directed workshops on web scraping and Kaggle for 100+ participants. Organized industrial visits, facilitating logistics for 30+ students.