About

AngelsWorks Hub is the engineering platform underneath everything else — the substrate every other module plugs into, and the machinery that turns specs into running modules. A multi-tier LLM gateway routes across 12+ providers with self-healing circuit-breaker fallback that has held through provider outages with zero downtime, alongside an RBAC model with trust scoring and a Mission Control surface that makes the whole thing observable rather than guesswork. A strangler-fig migration moved it module by module — without breaking the running thing.

ArtLens is a vision pipeline for art identification. Iconographic deduction over expert-curated subject variants, forced-choice VLM prompting, an image-prep stage that surfaces details no model sees in raw thumbnails, and a self-training loop that promotes confirmed observations into diagnostic elements after three independent corroborations.

Tenant Manager is an MCP-native SaaS for landlord operations. 48 MCP tools, full Belgian rental law as a first-class domain, and agents that can run the business as long as the human confirms.

In parallel: a separate, larger project is underway — the foundation layer for an agent-native application platform. Distinct from the three above. Quiet until it's ready.

At Electrabel he led the SharePoint development team, owned a 30,000-user collaboration platform through the SP2007 → SP2010 migration with production-deploy ownership, introduced continuous integration, and coached eight junior engineers — three written references on file. The rest was the same kind of work across regulated environments: life insurance at AG Insurance, retail banking at KBC. Between 2016 and 2018 he built the Contract Registration Tool at AG Insurance IS Life (via Aprico Consultants) on a legacy stack and tracked down a two-year-old production bug the team had learned to work around. Behind all of it, a long consulting career through his own company (Engelen INC, 2008—present), and BaaN ERP, CRM, and BI work earlier on across manufacturing and finance.

The point isn't the line items. It's that he learned how to ship and operate software inside organizations where just rewrite it wasn't a valid answer. That habit transfers.

Ideally with a team that owns its AI platform rather than buying SaaS abstractions, where decisions about routing, evaluation, prompts, agent boundaries, and data are visible and revisable rather than hidden behind a vendor's roadmap.

Concretely, what a strong fit looks like:

• A permanent, embedded seat on one team — full-time, direct employment (employer-of-record or consultancy partner both fine). The kind of role he stays in long enough to own the consequences of his decisions.
• Applied research teams that ship product — Anthropic, OpenAI, DeepMind, Mistral, and their EU analogues are squarely in scope. The thread that ties them together is that the work reaches users.
• A team that's past product-market fit, where the surface area is bigger than what one person can hold. He's built enough from scratch on his own time; the next step is to do it at a scale that needs more than him.
• A team worth spending years with. Compensation matters, but the multi-year fit matters more.

He writes small, commits often, and reads the code before changing it. He prefers making one decision well and writing it down to making three decisions vaguely and revising later. He'd rather route every LLM call through a single gateway and pay the integration tax than scatter provider keys across services, and instruments before adding features. He's comfortable in the boring parts — migrations, backups, observability, RBAC — because the interesting parts depend on them.

His consistent edge is lateral. He tends to find the cut from a different angle than the canonical approach — iNaturalist's research-grade verification model transposed onto art iconography in ArtLens, the strangler-fig refactor applied module-by-module in Hub, MCP-native rather than chat-bolted in Tenant Manager. Analytical from unusual angles is the through-line.

The case studies carry architecture sketches and the decisions behind them.