Melis Kilic

At QKare, I build production AI and data infrastructure for a B2B HR analytics SaaS platform and Luna AI, a TÜBİTAK 1507-funded recruiting R&D programme. I ship multi-agent analytics and coaching features with LangGraph and FastAPI, turning workforce data into conversational insights HR teams use daily.

I make AI quality measurable through automated evaluation pipelines, including an LLM-as-judge harness for simulated multi-turn conversations across five weighted dimensions. I also designed evidence-grounded CV extraction and validation workflows that reached a 99.58% evidence-grounded rate across 239 profiles.

I build governed data layers stakeholders can trust, from a 109-model dbt Core layer supporting 56 mart tables to PostgreSQL row-level security that prevents cross-tenant data access. My work includes traceable FastAPI gateways, OpenTelemetry hooks, Docker builds, and tests that protect data correctness on every pipeline change.

I also apply rigorous ML evaluation to healthcare AI, improving the Stroke Classification project's validation protocol and building models that achieved 98.0% F1 and 98.2% accuracy across 49,500 CT images. I'm motivated by systems where reliability, safe AI behavior, and verified data are treated as core product outcomes.