Hi, I'm
Mikhail Shytsko
Founder of Seedfast. Solutions & AI Engineer at Slovenská sporiteľňa, based in Bratislava, Slovakia. BS in Computer Science, Technical University of Košice (2025).
sporiteľňaSeedfast
Focused on creating value, not just writing code.
Shipping production systems where mistakes are expensive: agentic LLM applications and MCP agents at Slovenská sporiteľňa, Slovakia's largest retail bank, on a JVM/Python background in high-load services for 2M+ customers.
An AI-native platform generating large-scale synthetic datasets for compliance-sensitive industries — finance, pharma, medtech. It reads database schemas and produces realistic, relationship-aware data in under 3 minutes. Won 1st place at DDAccelerator Finals as the only Slovak startup competing across 9 countries.
Where I've shipped.
- Mapped the credit-review workflow from scratch with analysts and process owners, then replaced its manual multi-system lookup with an AI agent that summarizes and cross-references lending, collateral and limits spread across several internal banking systems.
- Got it approved for production in a regulated bank: a Koog agent sandboxed in its own virtual filesystem, reaching bank data only through a least-privilege, auditable MCP tool layer.
- Designed the agent's data access with strict data minimization and a controlled disable path — safe inside a regulated bank without altering the core approval workflow.
- Built change-detection that surfaces and explains material changes across products, limits, collateral and client data — so reviewers catch shifts they'd otherwise miss across systems.
- Shipped LLM pipelines for automated credit-data validation and risk scoring.
- Ran the LLM layer on Azure OpenAI and Anthropic models, traced and evaluated with LangSmith, LangFuse and Ragas to keep agent behaviour observable and measurable.
- Cut response times 85% for two high-load services serving 2M+ customers by taking sole ownership of inherited services and driving targeted performance optimisations.
- Built React and TypeScript features for a large micro-frontend banking app on in-house UI libraries.
- Doubled feature delivery by building a 3-person engineering team from scratch — from role definitions to offers, plus code review and retrospectives.
- Built an AI system that generates realistic, referentially correct data for any database in any locale and any domain — sparing teams days of manual data engineering per dataset.
- More than doubled output quality on held-out data with evaluation frameworks and automated correctness checks that made a non-deterministic LLM pipeline measurable.
- Shipped the Go CLI and MCP server that let engineers and AI agents run Seedfast in production and CI, plus the runbooks and setup guides behind them.
- Own the web app's onboarding, lifecycle emails and free Postgres tooling in TypeScript, React and Next.js, with funnel analytics showing where new users stall.
- Take the demos and technical calls myself, and stay with each team until the product runs in their own environment — often debugging setups I can't see.
- Run the containerised infrastructure behind the product, including the CI that builds and distributes the CLI.
- Built a GDPR-compliant backend from scratch on Spring WebFlux and GCP for a mental-health startup, encrypting patient data in real time with zero unauthorised reads.
- Took deploys from 12 minutes to 4 with no downtime by automating the GitLab CI/CD pipeline.
- Halved document-certification time by automating equipment certification.
- Accelerated product decisions ~25% with GraalPy trend analytics.
- Cut shift-approval time from 5 days to 2 through schedule optimisation.
- Shipped the secure Spring Boot REST API behind it, with OAuth2 / JWT authentication.
Selected work.
A mix of products, platforms and developer tooling — built to be owned, not just demoed.
Background & highlights.
Get In Touch
Open to collaborations, interesting problems, or just a friendly chat.