AI systems engineer · Co-founder, Interviewlary
I turn ambitious AI ideas into dependable products—from architecture and agent workflows to backend delivery, testing, and product decisions. Currently building Interviewlary in Istanbul.
Role-specific voice and text interview practice with scoring and actionable feedback.
01 · Selected work
Voice evaluation, token efficiency, agent reliability, and rapid multi-agent delivery—each case shows the build, my role, and the verification scope.
A live interview-practice product that adapts sessions to role, seniority, industry, and skills, then turns voice or text responses into a structured improvement plan.
Co-founder ownership across requirements, product flows, technical review, QA, positioning, and day-to-day delivery.
Visit the live Interviewlary productA TypeScript optimization layer for agent workloads with exact and semantic caching, prompt optimization, provider adapters, budget controls, and Model Context Protocol tooling.
92/92 repository tests passing. A controlled mock coding-agent A/B run showed 35.5% fewer billed tokens with identical output fidelity.
Review the TokensCache repositoryA defensive verification layer for autonomous coding agents, designed to catch fabrication, enforce evidence requirements, and fail closed when confidence is not earned.
Twenty-four adversarial evaluation scenarios cover fabricated completion, missing evidence, stale state, and contradictory outputs.
Inspect the Reliability Guard repositoryAn adversarial four-agent fact-checking prototype built for the Medipol META AI Hackathon, with streamed analysis and a live stage demonstration.
Designed, implemented, tested, and presented in a seven-hour sprint, with the scope narrowed to the fact-checking flow and a reliable live demonstration.
Explore the TruthNet repositoryBrowse all 16 distinct projects with live, public, development, and early-stage status shown before each link.
02 · Capabilities
I translate product intent into testable architecture, then follow the work through interface behavior, backend logic, persisted state, and release evidence.
Multi-agent coordination, real-time voice flows, memory boundaries, grounded retrieval, provider integration, and failure-mode design.
React and Next.js interfaces, TypeScript services, FastAPI backends, PostgreSQL, Supabase, and deployment workflows.
Deterministic checks, adversarial scenarios, persisted-state testing, and clear separation of prepared, verified, and deployed work.
Requirements, prioritization, user journeys, QA, positioning, launch planning, and cross-functional delivery decisions.
03 · Technical notes
Build notes state the evidence scope. Research notes use dated sources and avoid turning preview claims into permanent facts.
How exact and semantic caching were evaluated with 92 repository tests and a scoped mock coding-agent A/B comparison.
Read the TokensCache engineering noteA verification harness for fabricated completion, missing evidence, stale state, and contradictory coding-agent output.
Read the Reliability Guard noteHow scope control, adversarial agent roles, a streaming backend, and stage-demo constraints shaped the hackathon build.
Read the TruthNet build note
04 · About
I am a Computer Engineering student at Istanbul Medipol University and co-founder of Interviewlary. I use modern AI tooling aggressively, but I do not confuse speed with evidence: interfaces, backend behavior, tests, and deployment state all need to agree before I call work complete.
Useful systems and honest scope beat impressive-sounding claims.
I trace product intent through UI, API, logic, persistence, and release.
AI-assisted workflows increase iteration speed; verification protects the result.
I make risk, uncertainty, and next decisions easy for a team to see.
05 · Contact
I am open to AI engineering roles, internships, technical collaborations, and serious product conversations. The fastest route is email.