Engagements & track record.

A contract-intelligence platform serving governmental, defence, and infrastructure clients — multi-billion-dollar post-execution obligation programs where a single missed clause-driven deadline can trigger eight-figure liquidated damages. The platform’s background-job system ran on a legacy XMLRPC/TGScheduler jobserver that had become the primary reliability risk.

Gergely designed and shipped a managed Celery framework to replace it: 10,787 lines of code across 42 files, porting 50+ legacy jobs with multi-tenant context restoration, migrated live with no maintenance window. Two years on, he remains the runtime owner.

On the same platform, the core checkout-status workflow — the mechanism controlling document state across the system — was rewritten end to end: ten months, eight iterated versions, spanning a Python/SQLObject backend and ExtJS and Angular front ends. The work is a study in modernising inside a living legacy system: each version shipped, verified, and superseded without breaking the tenants running on it. Alongside it: recurring production incident command for DB-lock and stuck-job classes, including hot-fixing a production database lock-up.

A self-initiated security research sprint on a production platform produced four pre-disclosure reports in 30 days: a subdomain takeover; a CSRF weakness rooted in Mersenne-Twister PRNG predictability; an SMS-MFA-bypass account-takeover chain; and a leaked-credential audit. Each was written to pre-disclosure standard and framed for the organisation’s ISO 27001 processes — findings an external pentest had not surfaced.

Building AI into an engineering organisation, not a demo: a Claude Code slash-command suite running in production; an MCP evidence pipeline; and a 1,000-merge-request review-corpus analysis that identified the recurring feedback patterns and produced a pre-merge tooling proposal automating roughly one sixth of them. The underlying personal footprint on the platform: 200 merge requests at a 61.5% merge rate, 138 tickets, 17+ initiative branches owned end to end.

At Pulilab, work for WHO and UNICEF stakeholders: a Scrapy-based data-enrichment pipeline covering 20,000+ medical clinics; analytics over 100,000+ datapoints with city-, country-, and global-level aggregation, geolocation, and statistical outlier detection; a 1.5-million-word CMS; and — notably — an OpenAI-powered content-strategy system with A/B testing shipped in 2021–22, before ChatGPT existed.

Venturi Systems built AI-driven urban traffic optimisation. As co-founder and CEO, Gergely secured two B2G contracts through 30+ municipal meetings, optimised 16 intersections across 3 municipalities, and served as Product Owner for an external EPAM design team. The engineering core: distributed simulation of 3,000,000+ data points across 140 simulated years on GCP 96-core instances, with ML-accelerated optimisation. The formative years before that: four years of large-scale Magento eCommerce on a 130,000-product catalog, and 12+ freelance client projects delivered end to end — including a cross-platform PWA newspaper with subscriptions and payments.