Onat
Yaricilar.

Designed and shipped a learning-to-rank model for product search using click and conversion logs as training signal. Engineered lexical, behavioral, and catalog features; improved NDCG@10 by ~7% vs. the BM25 baseline. Served the model via a FastAPI re-ranking service with in-memory caching, keeping p95 latency under 50ms under production-like load. Wrote performance-critical feature processing components in C++.

Built the customer-facing AI backend before Linkup's $10M seed round. They went into the raise with a live product, not a slide deck. Shipped production agent workflows using CrewAI and custom state machines to automate lead enrichment, outbound, and onboarding for early U.S. customers.

Led AI integration across a sales automation platform serving 6,000+ customers as Head of AI. The Arya module was demoed live to investors and was central to Ekmob's ~$4M raise. Integrated directly with ERP systems for clients including KPMG and large insurance companies.

Co-founded and delivered 30+ production agentic and RAG systems for enterprise clients including EA Sports and Team Slate. Designed and evaluated RAG pipelines for accuracy and failure modes, ran A/B tests in production.

AI agents that automate document-heavy workflows for teams. Retrieval routes requests to the right policy, tools execute actions, and outputs are written back to the system of record.