At Programmers Force, I build production ML systems for Shufti’s identity-verification platform, including a lead-qualification engine that improved tier-1/2 precision and recall from 70% to 100% on a 198-case benchmark.
I shipped a production Model Context Protocol server with 20 verification and knowledge tools spanning document and face checks, AML, KYB, and eIDV. I also built hybrid search across 757 internal sales documents and reduced retrieval latency from 28.7s to 9.1s.
Previously at ML1, I fine-tuned Qwen3-VL-30B for document-defect classification, deployed an INT8 YOLOv8n MRZ detection service, and improved pipeline accuracy from 91% to 98%. I bring hands-on experience across computer vision, language models, evaluation harnesses, model serving, and MLOps.
