AI Product Engineer — I take systems from architecture to a working product: retrieval, computer vision, role-based workflows, tests, CI.
I came from the product and funding side: 6M RUB (~$70K) in non-dilutive funding across 5 projects, listed author on registered MedAI software. That is why I build with provenance, review workflows, and audit trails.
Portfolio · Email · LinkedIn · Repositories
Currently building: Procedra
An AI system that turns operator requests, enterprise documents, public standards, and video context into review-ready industrial work-instruction drafts — under mandatory expert approval. Sole developer.
- FastAPI/Pydantic application end to end, from generation schemas and validation to JSON, Markdown, and PDF export.
- Hybrid semantic/keyword retrieval with source-influence scoring, and durable video processing with keyframe extraction and frame-level vision analysis.
- Deterministic fallback that keeps the product fully usable without an external LLM.
- Enterprise layer: 6 roles, organizations, projects, permission matrix, instruction versioning, review workflow, execution checklists, audit trail.
- Security audit of my own system: found that adversarial context could be labelled as confirmed and score high. Introduced typed provenance, blocked model-supplied approval state, added explicit blockers for hazardous and poisoned-source cases.
- Quality gate: 406 passing tests, clean mypy across 46 files, Ruff, Docker Compose, CI, and a 15-scenario pack for reproducible evaluation.
Working paper on SSRN, Abstract ID 7026918.
| Area | What I work with |
|---|---|
| Engineering | Python, FastAPI, Pydantic, SQL, JavaScript / TypeScript, Docker, GitHub Actions, pytest, mypy, Ruff, SQLite, REST APIs |
| AI & data | RAG and hybrid retrieval, LLM integration, structured output with schema validation, deterministic fallback, generation quality evaluation, video computer vision |
| Product | Discovery, PRDs, MVP scoping, market research, competitor mapping, unit economics, grant applications · Figma, FlutterFlow |
- Took an AI decision-support tool for pediatric immunosuppressive therapy from concept to registered software in 18 months; listed as an author on the state registration certificate (MedAI).
- Secured 6M RUB (~$70K) in non-dilutive funding across 5 early-stage digital products in medtech, edtech, tourism, social impact, and consumer technology.
- Owned research and grant work at an early-stage product studio: market research, competitor mapping, and market sizing that teams used to decide what to launch and what to cut from MVP scope.
| Repository | What it shows |
|---|---|
| procedra-industrial-ai | AI system for review-ready industrial work instructions: retrieval, video analysis, roles, versioning, audit trail |
| medai-case-study | public AI product case study: architecture, process, privacy-aware product thinking, lessons learned |
| engineering-drawing-ocr | limited-code OCR case study for technical drawings, preprocessing, and custom recognizer experiments |
| portfolio | project portfolio, selected work, and professional overview |
- MSc, Computer Science and Engineering — South Ural State University, "Digital Ural" elite IT program · Sep 2025 – Present
- BSc, Architecture and Civil Engineering — South Ural State University · 2021 – 2025
- Finalist, "Startup as a Thesis" (SUSU) 2025 · Semi-finalist, MSU Startup Accelerator 2022 · JPMorganChase job simulations on Forage 2026
- Russian native · English C1
