This role moved my work from notebooks and side projects to production-grade systems running on real production data. I worked across the stack - Python backends, a real-time 3D interface and applied ML/NLP.

• Real-time 3D digital twin - a full-stack web app that mirrors the live state of a production/warehouse line in an interactive 3D scene in the browser, built in a two-person team (FastAPI · React · three.js).
• Responsible-AI recruitment pre-screening - a decision-support system for HR: the AI ranks candidates and surfaces evidence, but a human always makes the final call and no one is auto-rejected. PII is masked before it reaches any AI step, designed around the field's academic literature and KVKK/GDPR.
• Predictive maintenance (RUL) - the data and feature-engineering pipeline that estimates the Remaining Useful Life of an industrial blade from sensor data (~71k rows of real production data), preserving time-series structure.
• Agentic AI & LLM systems - studied production-grade open-source agent frameworks and LLM knowledge-base patterns: tool-calling loops, skills, long-term memory / RAG and multi-agent orchestration (prompt chaining, routing, orchestrator-workers, evaluator-optimizer), and brought them into my LLM-assisted, test-first workflow.