Portfolio
AI agents and automations we build with Claude Code, plus the AI security work behind them. Client names are anonymized where we do not have written permission to name them.
We run our own company on a 200+ component agent system built with Claude Code, so what we ship for clients is what we already operate in production.
Challenge
Run company operations end to end (admin, invoicing, marketing, monitoring) without a large team.
Approach
Specialized sub-agents, hooks that gate quality, MCP integrations, evaluation sets, audit trails, and automated tests.
Outcome
Runs 24/7 in production. The same patterns become your build.
Tech Stack
An online store running on Sylius needed a redesign, a technical audit, and extra functionality its own team could operate afterwards.
Challenge
The storefront needed to be rebuilt and extended without losing what already worked commercially.
Approach
Redesign on the existing Sylius platform, a technical audit of what was there, then the extra features on top.
Outcome
Ongoing engagement with the client.
Tech Stack
A fleet operator needed its vehicle and operations data in one place instead of scattered across spreadsheets and separate tools.
Challenge
Vehicle, route and cost data lived in disconnected files, so ordinary questions turned into manual assembly work.
Approach
A consolidated data platform with automated ingestion and reporting built on top of the sources the operator already had.
Outcome
Delivered and handed over to the client's own team.
Tech Stack
We built our own invoicing pipeline into Romania's national e-invoicing system, and we run every one of our own invoices through it.
Challenge
Every invoice has to reach the national system in a strict XML format. A failed submission is a compliance problem, not something you quietly retry tomorrow.
Approach
UBL 2.1 CIUS-RO XML generation, OAuth2 against the tax authority API, a circuit breaker for when the endpoint is down, and an offline queue so nothing is ever lost in transit.
Outcome
In production on our own invoices, with 181 of 183 automated tests green.
Tech Stack
How we assess external security and AI-threat posture before procurement or funding due diligence. Passive reconnaissance only, no active pentest.
Challenge
Know your exposure and AI-specific risks before someone else finds them.
Approach
DNS, web app, email security, code and data exposure, an AI and ML threat model, and EU AI Act posture.
Outcome
A 15 to 30 page report plus a prioritized remediation roadmap.
Tech Stack
How we take an AI system from unclear to audit-ready against the EU AI Act.
Challenge
Know if your system is in scope and what you must do before the deadlines.
Approach
Risk classification (Annex III and Article 50 transparency), gap analysis, documentation and audit-trail setup.
Outcome
A clear classification and an audit-ready action plan.
Tech Stack
Active collaboration with UTCB (Technical University of Civil Engineering Bucharest): a structured practice programme and internships, plus input from us on course content.
Challenge
Keep our methods current with research and build a talent pipeline.
Approach
Joint work on study programs and a structured internship.
Outcome
Ongoing public partnership.