I build data infrastructure for quantitative crypto research. Currently shipping two products:

Aperiodic — Professional market data aggregation for crypto. Microstructure, liquidity & order flow metrics — delivered as research-ready aggregates so you can run accurate backtests 100-1000x faster than raw L2 approaches, without petabytes of storage or months of infrastructure work. TrueOHLCV, execution-aware pricing, tick/volume/dollar bars, slippage metrics, order book aggregates — full universe, no survivorship bias.

Unravel Alpha — Cross-sectional alpha factors and multi-factor portfolios for systematic crypto traders. Point-in-time reconstruction, risk overlays, full transparency. We also manage external capital via separately managed accounts.

Professional-grade crypto market data aggregation. We handle the data plumbing so you don't have to.

Core metrics:

What makes it different:

• Not raw data — validated, exchange-normalized aggregates ready for research
• Exchange vs. local attribution — TrueOHLCV based on actual tradeable prices
• Execution-aware pricing — average fill prices at realistic position sizes
• Full universe coverage — no survivorship bias across all exchanges
• Enterprise-grade QA — standardized validation, point-in-time accuracy, documented data lineage

Featured:

Spectra — Our flagship licensable multi-factor portfolio.

Retail Flow — Cross-sectional factor that takes systematically contrarian positions against predictable retail herding behavior.

Open source tools:

unravel-client — Python client for the Unravel API. Retrieve real-time and historical factors, build portfolios, run backtests.

api-guide — Jupyter notebooks to get started. Portfolio construction, risk overlay integration, factor exploration.

crypto-predictive-risk-factors — Reference implementations for systematic crypto strategies using cross-sectional and alternative data.

Accelerating Spatial Analysis with Neural Networks (MSc Thesis, UCL Bartlett) — Proved that a Multilayer Perceptron can estimate Visibility Graph Analysis values without expensive graph computation. The trained network generates spatial configurations from VGA inputs and calculates neighbourhood size and clustering coefficients substantially faster than traditional methods, with negligible error. The system is space-generic — trained once, applicable universally. The implication: spatial analysis becomes interactive and real-time, enabling optimization procedures like genetic algorithms during the design process.

fold — Fast adaptive ML for time series. Composite models, online learning, temporal cross-validation. Built when I got tired of tools that assume stationarity.

krisi — Time series evaluation with PDF/web reporting. Metrics tracked over time, because a single RMSE tells you nothing about drift.

modular-pipelines — Multi-model ensembles and meta-model orchestration.

laplace-gnn-recommendation — Self-supervised graph neural network framework for recommendation via edge prediction on knowledge graphs.

deep-reinforcement-learning — DQN with prioritized experience replay, DDPG for continuous and multi-agent environments.

Aperiodic (Co-Founder & CTO) — Professional market data aggregation for crypto. Research-ready aggregates — TrueOHLCV, execution-aware pricing, tick/volume/dollar bars, order book metrics — so quant teams can run accurate backtests 100–1000x faster without petabytes of raw data. Python & Rust ingestion, TimescaleDB, Cloudflare Workers, React frontend. Serving data with sub-10s latency to systematic funds.

Unravel Alpha (Co-Founder & CTO) — Cross-sectional alpha factors and multi-factor portfolios for systematic crypto traders. Point-in-time reconstruction, risk overlays, full transparency. Also managing external capital via separately managed accounts.

Myalo — Quantitative research infrastructure for adaptive time series forecasting. Open-sourced fold and krisi from this work.

Health Venture Lab (Co-Founder & CEO) — EU-wide healthcare accelerator powered by GE Healthcare, in partnership with EIT Health. Faculty at MIT Linq Catalyst Europe.

UCL Bartlett (MSc Architectural Computation) — Genetic algorithms, morphogenetic simulation, L-systems, neural network–accelerated visibility graph analysis. Where I learned to think about emergent systems — it turns out markets are one.

Full-stack — Shipped products end-to-end in React/TypeScript, Python, Rust, Node.js. From SaaS platforms to VS Code extensions to data pipelines.

Berlin · Previously London & Budapest · English, German, Hungarian · Founders Pledge member

If you work in systematic trading, quantitative research, or crypto market microstructure — I'm always happy to talk. I follow back people who build interesting things.