CV

• Developed MomTest AI end-to-end: an agent loop using function calling and structured outputs, with strict guardrails to prevent leading-question bias in automated customer-discovery interviews.
• Integrated dynamic Markdown-based rule files for methodology enforcement; deployed and operated the production environment on AWS (EC2, S3, IAM, CloudWatch) and built unit/integration test suites.
• Owned model training and fine-tuning for GİB CAPTCHA OCR, raising recognition accuracy from 41% to 93%.

• Designed and built a database-agnostic Text2SQL agent end-to-end, from architecture through prompt engineering.
• Built a hybrid retrieval (dense + BM25) architecture with Qdrant (HNSW) for dynamic schema retrieval and selection.
• Reached ~95% accuracy on simple queries and ~85% on medium-complexity queries.
• Load-tested with Locust and monitored latency (p50/p95/p99) via Prometheus & Grafana.

• Designed and trained an edge-attentive residual CNN from scratch to denoise raw radar imagery.
• Implemented hybrid activation and loss functions to preserve high-frequency structural detail.
• Applied ensemble learning across multiple architectures and managed the end-to-end training pipeline.

Programming Languages

Machine Learning & Deep Learning

NLP, LLMs & Vector Search

Data & Analysis

MLOps, Cloud & Deployment