Current work
Vultax (opens vultax.com in a new tab)
Founder & Lead Engineer · 2025 — PresentVultax is a crypto market intelligence platform built on live market, transaction and blockchain data, which I am the sole architect of. The core of the work is treating market behaviour as something that develops through time rather than as a series of isolated observations.
Data infrastructure. I built the ingestion layer that collects and normalises streaming data across exchanges, trades, order books, prices, volume, blockchain transactions and external events. On top of that sits market-abuse detection, order-book analysis, whale and large-transfer monitoring, arbitrage discovery, market-quality scoring, wallet intelligence and event-driven market analysis.
Wash-trading detection is a good example. We look at several behaviours together: circular value movement, timing regularity, volume, price impact and the relationships between transactions. Writing a rule that flags something unusual is the easy half. The real problem is working out how much evidence each observation provides, and what it means when several weak signals arrive at once.
The scoring systems follow from that. Different measurements describe different parts of a market, they are not equally informative, and they are not always available. The system has to combine what it has without manufacturing certainty the data does not support.
Execution. The cross-exchange arbitrage and routing systems compare markets after fees, liquidity and execution constraints, rather than chasing the largest gap between two displayed prices.
Traceability. Large movements are tracked across both exchange and blockchain data, preserving chain, wallet, counterparty, exchange attribution, USD value and transaction hash, so any result can be followed back to the event that produced it.
AI layer. An analytical layer built over the platform's own market and research data rather than a general-purpose chatbot. Much of that work has been adversarial: finding where the reasoning fails — stale context, incorrect entity resolution, conclusions the data does not justify — and restructuring the system around the model to close those gaps.
Previously
Genius Sports
Football Trader · May — July 2024Priced live football markets in real time, across simultaneous fixtures — independent judgement calls made in seconds and defended afterwards, where a missed detail is immediately visible and immediately costly. Much of it was watching for price movement that did not match what was actually happening in the match.
What I work with
- Languages
- Python · TypeScript · JavaScript · SQL · Bash
- Backend
- FastAPI, Starlette and Uvicorn. SQLAlchemy and Alembic, asyncpg and psycopg, Pydantic, httpx and aiohttp. REST APIs, WebSocket services, background workers and scheduled jobs.
- Data stores
- PostgreSQL and TimescaleDB, Redis, ClickHouse, BigQuery, SQLite, S3-compatible object storage. Schema design, migrations, partitioning, retention and query tuning.
- Analysis
- pandas, NumPy, scikit-learn, PyArrow. Time-series and L2 order-book data, transaction graphs, statistical baselines, weighted scoring and confidence models.
- Frontend
- Next.js, React, TypeScript and Tailwind. TanStack Query, Lightweight Charts, MapLibre GL. Streaming interfaces over live market data.
- AI
- LLM-backed analysis built over first-party data: tool and retrieval design, evidence selection, and guardrails that stop the model asserting what the data does not support.
- Infrastructure
- Linux and systemd, nginx, Docker, Cloudflare, Vercel. Blue/green releases, pgBackRest backup and restore verification, Prometheus metrics, freshness guards and alerting.
- Sources
- Twenty-four exchanges and DEXs including Binance, Coinbase, Kraken, OKX, Bybit, KuCoin, Bitfinex, HTX, Uniswap and PancakeSwap. Polymarket and Kalshi, The Graph, Whale Alert, and on-chain data across Ethereum, Bitcoin, Solana and Polygon.
I am most interested in mathematics and statistics as engineering tools rather than abstract exercises: probability, distributions, optimisation and time-series behaviour.
What I am good at
Data is the part I keep returning to — taking a high-volume, continuously arriving, frequently messy stream of it, from market feeds, order books, blockchain transactions and external sources, and turning it into something structured enough to reason about.
Past that, the work becomes statistical. Deciding what a baseline should be, how far a deviation has to travel before it means anything, which signals are genuinely independent of each other, and how to combine them into a result that stays honest about its own uncertainty.
Most of the problems I enjoy start the same way: a lot of data, and a question that is much harder to answer than it first appears.
Blog
All postsEducation
Tallinn University of Technology
Expected 2027Bachelor of Engineering, IT Systems Development.