Vienna, Austria
Head of Enterprise, everii Group.
Twenty years in one codebase.
I've spent the last two decades building TEAMBOX, the ERP platform for service providers that's become the PSA market leader in DACH — starting as one of its earliest developers, now running enterprise integrations, CI/CD, and the team behind it. My current focus is agentic engineering: building AI-driven development workflows into a live, twenty-year-old codebase without breaking it.
20+ years, one productPHP & DevOps300+ enterprise clientsAgentic engineering focus
Work Highlights · July 10 – July 31, 2026
The latest work
Updated monthly, straight from the actual work — not a highlight reel.
*Dominated by a large, mostly-mechanical test-suite migration — hand-authored business logic is a smaller slice of that total.
Featured build
The test-suite migration, phase two: from migrated to actually trusted
Last cycle's migration got every module speaking one test format. This phase went after what that migration didn't fix on its own: parts of the suite were still partially skipped and non-blocking in CI, quietly hiding bugs of their own — 230 commits over three weeks turned that into a required, fully green gate.
- Transactional isolation, not manual cleanup. Built a PHPUnit extension that wraps every test in a rolled-back transaction, replacing thousands of lines of hand-written cleanup code across the whole suite.
- Un-skipping is how you find out what's actually broken. Eighteen previously-disabled tests came back online — and several had been quietly hiding real bugs, including a database reconnect that silently dropped the session timezone on every request.
- Order-dependent tests, exposed. A transaction-savepoint depth leak on skipped tests, and tests that only ever passed because of the order they ran in — both invisible until the isolation work forced them to run standalone.
- A gate, not a suggestion. The suite moved from non-blocking to a required CI check — it now has to stay green, not just exist.
The migration itself moved the format; this phase is what made the result something you can actually rely on — a suite that fails loudly instead of one that quietly stops meaning anything.
Read the full writeup →Featured build
Reworked client numbering end-to-end to support alphanumeric IDs — while it was still moving underneath
An enterprise client running SAP alongside TEAMBOX needed the same alphanumeric client codes in both systems — a change that touches nearly everything client-numbering ever touches: two frontend stacks, a decades-old payment-reference format, and the assumption, baked in for years, that a client ID is just a number.
- Moved the field once the domain model demanded it. The new external-ID field started on the company record, then moved to the client record mid-build once that became the obviously correct owner — the kind of correction that's cheap early and expensive if you wait.
- Decoupled the parts that used to be tangled together. A new per-client job-number pattern engine separates project numbering and payment references from the new external-ID format, instead of leaving them implicitly coupled to "the client number is numeric."
- Immutable once set, on purpose. The external ID locks after creation to guarantee uniqueness — a constraint added deliberately, not a limitation discovered later.
- Caught its own regressions before anyone else did. A literal external ID of "0" was being treated as unset, and client search silently broke the moment IDs stopped being purely numeric — both found and fixed in review, before either could reach a real client.
Shipped behind its feature flag for a staged rollout, with a permanent end-to-end test suite added alongside it. The workflow behind how it actually got built — implementer, reviewer, and tester run as separate, independently-checked roles — is written up in full in Agentic Strategies →.
Read the full writeup →Featured build
Enforcing one path for schema changes surfaced a run of bugs that would have broken real deployments
Making a proper schema-migration tool the only way schema changes happen — no more parallel legacy upgrade paths — turned up problems that had gone undetected for years, the kind any twenty-year-old codebase accumulates.
- A flag that had never worked. The legacy upgrade tool's CLI flag parser had no working long-form flag at all — a bug invisible for as long as everyone happened to use the short form.
- A crash on every real run. One of its flags crashed on every actual execution — not an edge case, the normal path.
- A subtle auth mismatch. CI database authentication was failing from a password-decoding mismatch between the old credential convention and a newer one — the kind of gap that only shows up once both conventions exist at the same time.
- Cleanup with a paper trail. Fifteen historical legacy upgrade directories deleted once confirmed superseded, and the migration strategy now lives as an architecture decision record that gates schema-change review going forward — not just tribal knowledge.
None of these bugs were the point of the migration — they were what the migration had to survive finding, on its way to becoming the one real path for schema changes.
Read the full writeup →Agentic Strategies
How I actually build and work with AI agents
Long-form writeups on the methodology — scoping, review discipline, verification, documentation architecture — for anyone doing this kind of work themselves, not just a list of what got built.
Agentic Strategies
Context, Decisions, and Guardrails: Why Agents (and New Hires) Shouldn't Re-Derive Your Codebase From Scratch
Every fresh AI agent session on a legacy codebase faces the same choice: re-analyze everything from the code alone, or read what's already been decided. One canonical guide, read consistently by three different AI tools, and a strict split between what's being asked for, what's been decided and why, and how a specific piece of work is scoped.
Read the full writeup →Agentic Strategies
Instructions, Style, and Memory: How I Keep Several AI-Assisted Projects Consistent
Running several AI-assisted projects at once creates a specific failure mode: each one drifts its own way, and every session re-explains context the last one already worked out. Persistent per-project instructions, one shared style guide, and a memory system that carries lessons across sessions — the scaffolding that stops it.
Read the full writeup →Agentic Strategies
Implementer, Reviewer, Tester: A Workflow for Shipping Features Fast with Claude Code
Recently shipped a full production feature — new client/project number handling with edge cases nobody had scoped up front — in meaningfully less time than the "cheapest possible version" would have taken, using Claude Code as implementer, reviewer, and tester in separate, deliberately isolated roles.
Read the full writeup →Personal AI Research · Proof of Concept, July 2026
Outside TEAMBOX: independent AI research builds
A way to stress-test what Gemini can actually do on real, messy problems instead of toy demos — built independently of TEAMBOX, in Google AI Studio. Keeping both of these consistent with each other, and with the rest of this site, runs on the practice covered in Agentic Strategies →.
Personal AI Research
OSINT Real-Time Tracker — multilingual conflict monitoring, built solo with Gemini
A maps-based dashboard that monitors global conflict events close to real time. It takes open-source reports — social posts, wire dispatches, radio transcripts — in whatever language they were written, and turns them into structured, geolocated, credibility-scored events a researcher can actually use.
- Multilingual intake, no separate translation step. Paste raw source text in Ukrainian, Arabic, French, Hebrew, or English and Gemini handles extraction directly — entities, location, and language nuance in one pass, not a translate-then-analyze pipeline.
- NLP-based geolocation. Gemini resolves place names and context clues into exact coordinates, with the reasoning behind each resolution shown, not just a pin dropped on a map.
- Sentiment and tension scoring beyond positive/negative. Each report gets a sentiment score, a 1–10 tension score, a dominant emotion, and a read on native-language slang and nuance that keyword-based tools miss entirely.
- A transparent, five-parameter credibility grade. Every event is scored 0–100 across source reputation, cross-referencing, evidence level, contextual plausibility, and freshness — with the reasoning shown, not a black-box trust score.
- An autonomous harvester, not just a manual tool. A background scheduler runs the same search-grounded pipeline once an hour, unattended, aimed at a fixed set of named authoritative sources — ISW for Ukraine frontline reporting, UKMTO for Red Sea maritime alerts, IAEA for nuclear-safety monitoring at Zaporizhzhia, CISA and Mandiant for cyber incidents — rather than an open-ended crawl, with every harvest and its full audit trail persisted to Firestore.
This is a personal research prototype, not a production intelligence product — the live feed uses Gemini search grounding against open sources, with a documented offline fallback dataset for when that harvest fails. Built end-to-end (React, Express, Gemini API, Firestore) in Google AI Studio and deployed to Cloud Run.
View live app → Read the full writeup →Personal AI Research
Crimes of the 1% — a structured, sourced database of financial crime and litigation among the ultra-wealthy
A database that catalogs corporate and executive-level financial crime — fraud, market manipulation, tax evasion, state capture — against the actual enforcement outcome on record: criminal conviction, settlement, deferred prosecution, or litigation still in progress. Built the same way as the OSINT tracker: solo, in Google AI Studio, testing whether an LLM can hold a large, sourced, structured dataset together instead of just summarizing text.
- Sourced by design, not by afterthought. Every case record carries one or more citations classified by document type — DOJ filing, SEC order, court judgment, EPA record, ICIJ archive — instead of a single unlabeled link.
- Beyond a dollar figure. Each entry converts a financial penalty into a systemic-impact score and a "worker-wage-equivalent" figure — how many years of an average worker's pay the penalty represents — and a companion inequality dashboard turns the dataset into charts by sector and crime type.
- Two Gemini-powered lenses on the same data. An AI Fact-Finder modal answers ad-hoc questions against the dataset, and an AI Patterns modal surfaces cross-case trends — which sectors recur, which enforcement types are most common — the same "structured reasoning over messy real data" test as the OSINT build.
- A submission path, not a closed dataset. A case-submission form lets anyone add a new entry directly — published immediately with an "AI-generated, unreviewed" disclosure badge, so the dataset isn't limited to whatever I find on my own research pass.
Early-stage research build — React, TypeScript and Firebase/Firestore, seeded with an initial dataset rather than a continuously fact-checked ledger. Treat individual case entries as a starting point for your own verification, not a settled record.
View live app → Read the full writeup →