Portfolio
We don’t build generic wrappers. We engineer proprietary ML infrastructures that solve mission-critical bottlenecks across highly regulated, data-intensive industries.
& Risk Intelligence
Credit Scoring
Alternative lenders lose millions annually to information asymmetry. Our neural infrastructure autonomously ingests JPK tax files and bank statement histories — detecting hidden payment bottlenecks 90–120 days before actual delinquency.
Beyond financial statements, the system integrates social media behavioral analysis and open-source intelligence, cross-referencing public digital footprints with financial patterns. A custom LLM layer interprets unstructured text signals and synthesises them with our ML scoring model — producing a multi-dimensional risk profile that traditional bureau data cannot replicate.
Credit Risk Scoring
Built for large-scale institutional underwriting desks. Synthesises macroscopic market indicators — sector spreads, leverage cycles, counterparty exposure — with micro-transactional behavioral data.
Risk departments can safely expand credit boxes without increasing portfolio NPL rates. Every decision is explainable, auditable, and regulatory-compliant from day one.
Scenario Engine
Built as a core product layer, not an analytical add-on. Risk and investment teams work with the exact same synthetic dataset under fundamentally different financial realities — stress, conservative, aggressive, or market-driven. Assumptions are embedded directly into the financial logic that governs how portfolios behave, not applied as cosmetic adjustments.
The engine reshapes distributions, dependencies, and time dynamics according to explicit scenario definitions. Cash flows, default behavior, recovery paths, and portfolio volatility respond consistently — turning scenario analysis from a presentation exercise into a production-level decision tool.
& Energy Markets
Time-Series Generator
Engineered for energy traders, quant analysts, and HFT/hedging desks who need what historical data cannot provide. Our proprietary pipeline deploys TimeGAN, Copula models, and GARCH processes to faithfully replicate extreme TTF contract volatility — including autocorrelation regimes, European gas storage dynamics, and precise weather anomaly impacts on spot pricing.
Algorithmic trading desks stress-test execution and hedging strategies against statistically valid Black Swan scenarios at scale — without waiting for reality to deliver them.
& Healthcare Diagnostics
Coloplast Heylo
Medical hardware shouldn’t just react to problems — it should predict them. We took our battle-tested FinTech anomaly detection engine and applied it to Coloplast Heylo smart ostomy sensors. The result is the Intelligence Layer — an advanced Software as a Medical Device (SaMD) overlay.
Instead of a stressful, binary „leak detected” alarm after the damage is done, we use Edge AI and Synthetic Data to give patients real-time predictive intelligence. Zero hardware modifications. 100% local processing. Absolute privacy.
Instead of a post-failure alarm, patients see a live countdown — „Safe. Leak in ~45 min” — giving them time to act with dignity and confidence, not panic.
Continuous skin condition monitoring derived from sensor behavioral patterns. Early deterioration flagged before clinical symptoms appear.
Integration with Apple Health correlates physical activity and daily routines with sensor performance — actionable lifestyle insights for patients and clinicians.
Detection
A computer vision diagnostic system identifying malaria parasites directly from microscopic blood smear imagery. Engineered for low-resource clinical settings — dramatically reducing manual screening time while maintaining extreme precision through a custom CNN trained on synthetic augmented datasets.
& Computer Vision
Debris Detection
Space debris compounds with every launch — a catastrophic threat to all orbital infrastructure. Hawkeye-Zero classifies 11 distinct debris classes across static imagery, video feeds, and live telemetry streams, built on YOLO 11 for maximum inference speed.
Delivered as a clean Python tool for frictionless API integration into aerospace operations platforms, mission control systems, and satellite management infrastructure.
& Sports Tech
& Race Telemetry Engine
UltraKotlina is one of Poland’s most prestigious ultra-running events — recognised across the entire national running community. We are building a high-availability mobile application that processes live runner telemetry from the field: dynamic finish-time estimations, per-segment pacing strategies, and checkpoint logistics in real-time under extreme remote conditions with intermittent connectivity.
Beyond the product: UltraKotlina is actively pursuing entry into an international ultra-running world series — the equivalent of Formula 1 for trail running. National visibility across Poland’s entire endurance sports community, and a gateway into the international competitive circuit.
Ultra-Running World Series
— equivalent of Formula 1
for trail running.
& Consumer AI
Vehicle Cost Analyzer
Built entirely from scratch — Motobilans predicts real maintenance, fuel, and total ownership costs for any vehicle, helping drivers make smarter financial decisions before purchase or import.
Three specialised AI modules: AI Mechanic for diagnostics, AI Importer for complex international logistics, and AI Calculator for total cost of ownership modelling — integrated with carVertical, OpenAI, and Stripe.