Machine learning engineer
AKK company
Jun 2026 — PresentBuilt an OCR + LLM pipeline that extracts scoring features from scanned RU/KZ loan documents (bank statements, financial reports, land acts, crop-yield certificates), reaching 97.3% field-level accuracy on a hand-verified ground truth of 3,000+ values. Developed an LLM agent that routes corporate correspondence to the responsible executive with 87–91% accuracy (two-stage design, +25 pts over a single-prompt baseline); migrated it to an on-prem model with 82.8% accuracy for data-residency compliance Shipped both systems to production as FastAPI microservices (Docker, CI/CD to Kubernetes), with