Where theory meets messy reality.
PROFILE
PRAGUE, CZ
AGE 12 → 2026
20+ YRS SHIPPING CODE · CTU & CAS UTIA
I started coding at 12 and built my first paid website at 14. After years of developing full-stack applications as a freelancer, I studied Mathematical Engineering at CTU Prague and spent a decade researching reinforcement learning, Bayesian methods, and statistical decision-making at the Czech Academy of Sciences.
THE PRODUCTION GAP
Models often break when they meet production: sparse data, temporal leakage, shifting distributions, or optimizing a metric that ignores business economics. Because I build both the mathematical models and the surrounding software, I own the full problem. The objective is never a complicated model for its own sake; it is a reliable system that makes a measurable difference.
SELECTED WORK
Systems & Applied Models
Five production deployments, open-source engines, and applied models built end to end.
+5%
GROSS PROFIT LIFT
40%
HIT@3 ON SPARSE BASKETS
COMMERCIAL ML · OPTIMIFICA
Recommendations from transaction data, not clickstreams.
A CPU-only statistical recommendation engine built strictly from POS receipts, deployed across in-store self-scanners, cash desks, and e-shops for two retail chains and validated with a major supermarket.
INDEPENDENT SAAS · LIVE PRODUCT
Elevist: AI that connects goals, daily work, and results.
An execution system built on a relational PostgreSQL context graph that sorts daily tasks by economic leverage (
,000/hr vs. 0/hr) and runs an interactive Friday reality audit.OPEN-SOURCE AI INFRASTRUCTURE · APACHE 2.0
ClueState: Evidence-backed state for AI applications.
Separates probabilistic LLM claim extraction from deterministic state management, enforcing strict schemas and verifying character-offset source quotes chronologically.
DEVELOPER PLATFORM & COMPILER
CoevoApps: One project, compiled for web and mobile.
A low-code platform and compiler with a built-in MCP server that turns one project specification into standalone SvelteKit web apps and React Native mobile apps.
APPLIED RESEARCH · AZORES (2025 - 2026)
AI4PORTS: Underwater infrastructure inspection.
Computer vision and acoustic sonar fusion developed with the IS2E team in Portugal to detect and grade structural anomalies from robotic ROV feeds in turbid marine water.
ENGAGEMENT MODELS
Three ways we can work together.
Direct collaboration with one technical owner from problem definition to deployed software.
Recommendation Systems
Predictive & Statistical Modeling
Model Evaluation & Audits
Structured LLM Pipelines
Computer Vision
8 TO 20 HRS / WEEK
Fractional Senior ML Engineer
For teams that need senior ML expertise and implementation without another full-time hire. I work alongside your team to design, evaluate, and ship complex ML features, recommenders, and AI pipelines across Python, PyTorch, databases, and APIs.
2-WEEK FIXED SPRINT
ML Audit & Feasibility Sprint
For teams validating a new AI initiative or diagnosing why an existing model underperforms on real data. In two weeks, I review your data, assumptions, metrics, and architecture, delivering a written diagnosis, a reproducible statistical baseline, and prioritized next steps.
DEFINED PROJECT SCOPE
Prototype to Production
For companies with usable operational data and a defined problem to solve. I take end-to-end ownership from problem formulation and statistical modeling through database design, APIs, and user interface, handing over a tested, maintainable system.
EXECUTION PROTOCOL
One technical owner. Zero handoffs.
From raw data inspection to deployed production code.
UNDERSTAND
Clarify the business decision, inspect the raw data, and define what success actually looks like.
VALIDATE
Test assumptions, check for leakage or sparsity, and build a simple statistical baseline before adding complexity.
BUILD
Develop the model and the surrounding database, API, and interface in visible, working increments.
DELIVER
Deploy to production, verify against live metrics, and hand over clean, documented code your team can maintain.
MATHEMATICAL ENGINEERING · CTU PRAGUE · CZECH ACADEMY OF SCIENCES
Research & Publications
EDUCATION
M.Sc. & B.Sc. Mathematical Engineering
CTU Prague (Magna cum laude, GPA 3.97). Ph.D. thesis submitted 2026.
RECOGNITION
UTIA Best Junior Paper Award (2025)
Work presented at NeurIPS, ICANN, and ICCAIRO.
INTERNATIONAL R&D
6 R&D Projects
Contributed to 1 EU Horizon, 2 international, and 3 Czech national projects.
2026
FEATURED JOURNAL PAPER · IEEE ACCESS
N-to-1 Knowledge Transfer in Reinforcement Learning via Adaptive Q-Function Selection
Marko Ruman · Published in IEEE Access (2026).
PRACTICAL APPLICATION:
How to make models adapt to new customer segments or product categories when you have almost zero historical data.
SELECTED PEER-REVIEWED PAPERS
2024 · IEEE ACCESS
DOI ↗Knowledge Transfer in Deep Reinforcement Learning
Practical application: Cross-domain transfer learning to accelerate model training in data-scarce environments.
2021 · WILEY (IJACSP)
DOI ↗Mixture Ratio Modeling of Dynamic Systems
Practical application: Detecting hidden regime shifts and behavioral changes in noisy real-world time-series data.
2016 · SPRINGER LNAI
DOI ↗Adaptive Proposer & Deliberation-Aware Responder in Multi-Proposer Ultimatum Game
Practical application: Behavioral game theory and multi-agent interaction modeling under uncertain payoffs.
WORKING PHILOSOPHY
A prediction is not a product.
A model is not a solution.
Make it useful.
START A CONVERSATION
Have a data or AI problem worth solving?
I work remotely with founders and engineering teams across Europe, the UK, and the US. Tell me what you are building, what data or systems you have, and where you need help. I will reply personally within one business day.
DIRECT CONTACT
Based in Prague, Czech Republic · Working remotely across EU, UK, and US timezones.
START A CONVERSATION
YOUR NAME & COMPANY
EMAIL ADDRESS *
WHAT ARE YOU TRYING TO BUILD OR SOLVE? *
Direct reply from Marko within 1 business day.