March 2026 — present · 8 months
Technical Product Manager
Severstal · Metals, manufacturing, and mining
SuperSlide is an AI product that creates corporate presentations using an internal LLM within the company’s secure environment. I took it from discovery and independent development through pilots, production rollout, and scaling.
- Reduced the time to create a first presentation draft from several hours to about one minute. Ran pilots with groups of 50–100 people and three rollout waves of roughly 1,000 users each; the potential audience is about 50,000 employees.
- Built a working version with AI agents in two weeks. The solution passed internal selection and was chosen for further development.
- Conducted interviews, surveys, and process analysis; validated hypotheses with working product versions.
- Defined positioning, roadmap, backlog, and prioritization based on user impact, effort, and technical risk.
- Defined product, user, and technical metrics: usage, retention, share of outputs requiring no manual edits, speed, latency, and reliability.
- Designed the LLM pipeline: the model creates the structure and selects slide types, while a deterministic engine calculates layout and generates an editable PPTX.
- Established AI quality evaluation using reference examples, user and expert reviews, and error and hallucination analysis.
- Wrote user stories and acceptance criteria, assigned work, and accepted delivery. Worked with engineers to adapt the product to the corporate stack and Kubernetes.
- Presented the product to executives and the steering committee, aligned security, infrastructure, and resources, and organized training and support. Built production code with AI agents and independent agent review.
Stack: TypeScript, Python, Next.js, PostgreSQL, Qdrant, Temporal, MCP, Docker, CI/CD, Claude Code
Case studies: SuperSlide · Agentic Experience