Oleg K.

I build AI products and the systems underneath them - from the first architecture call to production traffic at scale.

I do two things that rarely come in one person: design and ship an agentic product end-to-end, and keep the infrastructure under it from falling over, leaking data, or outspending its revenue. 17 years on the second part; the first is where I spend most of my time now.

What I do:

AI products from zero to production. Multi-provider LLM routing with fallbacks, usage and credit metering, agent and tool design, evals, guardrails, tracing, cost-per-task budgets. Model-agnostic: Claude, OpenAI, Gemini, open-weight and self-hosted.
Built: AI SaaS from scratch - 52+ models behind one interface, expert personas as prompt + model + tool chains, image and video pipelines, web and mobile apps. 60K users. 200K chat completions and 1000+ image/video generations a day.

Platforms that stay up under load. Event-driven microservices, Kafka, ECS Fargate, RDS, ElastiCache, Terraform/CDK. I treat architecture, reliability and AWS cost as one problem, not three.
Built: AdTech SaaS - an 18-microservice AWS platform that has analysed $17B+ in ad spend across Google, Microsoft and Meta Ads. Demand-driven autoscaling, graceful degradation and end-to-end observability for report collection from 200K concurrent users; shipped TikTok and InMobi integrations on top.

Agents that replace real operations. Automating manual workflows where agents actually pay off - and saying no where they don't.
Built: internal tooling for a Microsoft Ads partner team supporting 7,000+ advertising accounts. automated account linking, spend reporting and campaign-metric aggregation, replacing manual reconciliation across per-account reports.
AI-native engineering process. I run delivery on agents - multi-agent Claude Code workflows, parallel review agents, custom harnesses - and set this up for the teams I lead, so a small team ships like a large one.

Product ownership. Architecture and scope calls, hiring and leading engineers, owning what happens in production.

Track record:
✓ AI SaaS: 0 → production multi-model platform, web + mobile
✓ AdTech SaaS: 18-microservice AWS platform, $17B+ ad spend analysed
✓ AI-driven EdTech platform: MVP → 100K+ users, profitable
✓ API latency 2s → 180ms; zero-downtime migration of a 70K-user monolith to event-driven services
✓ Payment infrastructure at $1M+/yr, PCI DSS compliant

Stack: TypeScript (NestJS, Next.js), Python (FastAPI), PHP (Laravel), Postgres/Supabase/pgvector, AWS (ECS Fargate, RDS, ElastiCache, SQS), Kafka, Terraform, CDK, MCP, (React Native/Expo).