About

Head of AI at Zoral, where I’ve worked since 2004 and led AI R&D since 2013. I’ve been writing software since 1995.

I turn open-ended problems into production systems, from early research and architecture through implementation. My work spans machine learning and NLP, language and rule-engine design, large-scale web data extraction, and high-performance financial software.

AI & ML: neural networks, genetic and evolutionary algorithms, NLP, NER, classification, language identification, sentiment analysis, expert and rule-based systems, fuzzy logic

Language design: DSLs, rule engines, interpreters and VMs (RL3, Kavun)

Languages: Go, Python, C++; also C, R, Lua, Prolog, J

Data: PostgreSQL, TigerBeetle, Redis, Elasticsearch/Lucene, Neo4j, BigQuery, OLAP

Systems: distributed computing (NATS, MPI, OpenMP), low-allocation performance engineering, Docker, Kubernetes, AWS, GCP

Selected Work

  • Core banking platform — architected and built a high-performance, DSL-based core banking platform.
  • aiHit — led a team of 20+ engineers and 10+ linguists building an automated company-data platform using web, news and social sources. Owned project delivery, architecture and R&D, and implemented core components including the intelligent crawler, website categorization and language detection.
  • RL3 — designed the language and engine for rule-based information extraction, named-entity recognition and categorization (examples).
  • AI R&D — deep learning, automated underwriting, self-learning systems and predictive data-quality management for financial services.
  • Applied research — evolutionary (genetic-algorithm) car design, OLAP-based data-quality analysis, traffic and location prediction.

Open Source

  • Kavun — a lightweight, high-performance embeddable scripting language for Go, with first-class records, lambda-based pipelines and a sandboxable VM.
  • Decimal arithmetic and financial tooling for Go:
    • Dec128 zero-dependency 128-bit fixed-point decimal with exact SQL NUMERIC / PostgreSQL semantics,
    • Fin128 deterministic financial mathematics,
    • PgxDec128 zero-allocation pgx v5 codec.
  • nats-mutex — distributed locking on NATS JetStream.
  • Performance building blocks — slab, refpool, densemap, gobu.

Side Projects

  • HappyGrumpy — Twitter sentiment analysis service (now offline): idea, backend, frontend and mobile apps. Featured on FOX 35 Orlando and FOX 5 New York.