I build end‑to‑end data & AI products.
I lead the data team at CGTrader, shipping applied AI into production — visual search, pricing models, LLM agents. Ten years end to end, currently focused on applied LLMs and agentic AI.
I'm a Production Engineer who became a Data Scientist. Ten years building end-to-end — warehouses, pipelines, ML systems, and most recently LLM-orchestrated agents in production. I'm drawn to problems where the answer isn't obvious.
At CGTrader, I lead the data team across engineering, science, and analytics for the Marketplace, CGDream (browser-based 3D AI generator), and Modelry. I architected the company's 290+ model dbt warehouse on BigQuery and shipped applied AI into production — CLIP-based visual search for 3D models, a real-time price-suggestion service for sellers, and two LangGraph + Claude agents that handle internal data requests end-to-end.
Before that, I led Data Science and Machine Learning at Reccodo, where the team owned Recommendation and Personalised Search end-to-end. We moved recommendations off heuristic and matrix-factorisation baselines onto embedding-based retrieval with learning-to-rank reranking — two-tower architecture and online serving — and built the MLOps backbone, so the team could ship rather than wrangle infrastructure.
On the side, I'm co-founding Sinequity, the first M&A for SMBs platform in Greece, built around an LLM-orchestrated valuation engine. I also sit on ITeQ's board as a Data & AI advisor.
I work hands-on alongside the team. I like problems where the answer isn't obvious, and teams that deliver innovative solutions.
Selected roles, last ten years.
I lead the data team across engineering, science, and analytics for CGTrader's Marketplace, CGDream (browser-based 3D AI generator), and Modelry. Player-coach: hands-on with code review, system design, and modelling alongside hiring, mentoring, and stakeholder work with product and the C-suite.
- Architected the company's 272-model dbt warehouse on BigQuery
- Shipped CLIP-based visual search for 3D models, and a real-time price suggestion service for sellers
- Built two LangGraph + Claude agents that handle internal data requests end to end
A marketplace for buying and selling Greek SMBs, built around an LLM-orchestrated valuation engine, agentic workflows for enrichment, and a data warehouse over the Greek business registry. Ongoing involvement is oversight-level while a co-founder runs operations.
Advise management on data and AI strategy across the company's pharmacy network and SaaS platforms.
Led the data science and ML team owning recommendation and personalised search systems end to end — from training data and architecture to deployment, monitoring, and online inference.
- Moved recommendations off heuristic and matrix-factorisation baselines onto embedding-based retrieval with learning-to-rank reranking; two-tower architecture, online serving
- Built the MLOps backbone: feature pipelines, model registry, training and deployment automation, evaluation harnesses
Built and shipped the personalised search engine: ElasticSearch retrieval with a personalisation layer that tailored results per user via product embeddings and Redis-backed user profile vectors. Maintained the deployed ML estate and rebuilt hot paths into vectorised implementations to keep inference fast as the catalogue grew.
Mentored students through the Data Analyst Nanodegree: 1:1 guidance on projects, code review, and translating advanced concepts for varied levels of audience sophistication.
Delivered packet-level network analytics at carrier scale for Tier-1 APAC mobile operators (Indosat and Three) — covering network performance optimisation, subscriber behaviour and application usage analytics, segmentation, churn, and ARPU use cases. Owned the full project lifecycle alongside two other PMs sharing a 6–7 engineer team.