I build production AI infrastructure: multi-model orchestration, agentic systems, and the plumbing that makes LLMs cheap and reliable at scale.
- JAI Portal — AI marketplace unifying 500+ models. Built solo; 200+ countries, ~$60K/month.
- Chat.JAIPortal — 100+ LLMs in one product: side-by-side Chat Arena, an MCP agent that builds apps, 21+ integrations.
- Libranotes — meeting bots for Meet, Zoom and Teams; diarized transcripts in 32 languages; agentic RAG with citations. Built solo.
- RL for games — custom RL model serving real-time actions to an Unreal Engine 5 game; core of a ~30M TL TÜBİTAK R&D project.
Now building → argrouter — an LLM router that picks the model and reasoning effort per query from accuracy and real billed cost. Submitted to the RouterArena router leaderboard (under review).
50+ internal agents and pipelines
Game live-ops
- Live-ops automation — scheduled, multi-stage pipelines that run campaigns, battle-pass seasons, missions, offers and timed pricing end to end, with translation and cache invalidation
- Trading-card art pipeline — image-model generation constrained by rarity and numbering rules
Growth intelligence
- ASO intelligence — App Store search, autocomplete, reviews and charts, snapshotted into time series
- Early ad-performance prediction — hourly Meta Ads features feeding a model that calls winners and losers hours after launch
- Cohort alerting — attribution cohorts in, anomaly alerts out, with an LLM-written brief
- Trend and signal tooling — TikTok trend studio, social signal reports, PR media monitoring
Agents in operations
- Talent intelligence — embedding search with LLM reranking over candidate profiles
- Inbox triage — LLM classification of inbound mail, Slack escalation, scheduled follow-ups
- Creative operations — multi-account creative upload across ad platforms; resume-safe async generation pipelines
Python · FastAPI · PyTorch · LLM orchestration · RAG · MCP · speech pipelines · Docker — LinkedIn