Machine Learning Engineer
BTS Digital
Sep 2025 — PresentProgressed from an ML internship to owning production anti-fraud systems: face liveness, which cut HTER from 5.0% to 1.3%, and identity-document OCR, which raised accuracy to 95% and cut latency by 30%.
• Fine-tune vision-language backbones for face anti-spoofing against print, replay, mask and deepfake attacks, cutting HTER from 5.0% to 1.3% (FAR 0.12%) in production. • Built face detection and anti-spoofing pipelines with RetinaFace, PyTorch and OpenCV for automated content verification, a task closely aligned with automated moderation. • Ran transfer-learning experiments on deepfake video datasets, improving robustness on out-of-distribution samples. • Own an end-to-end OCR system for identity documents (IDs, passports)