I engineeradoption.
Interfaces between intelligent systems and the people who use them — protocols, products, and teams. This site is a digital twin of my work: every moving element maps to live state.
System map
projects and domains as a typed graph — hover to trace, click to inspectSystems connect to the domains they exercise. The real site renders this from the same RDF graph that powers my digital-twin work — the map is queryable, not decorative.
Selected systems
the live ones wear their own colors — full case studies behind eachSemantic digital twin for vehicle telemetry
Hybrid trajectory prediction and anomaly scoring over GPS streams, backed by a SPARQL-queryable SAREF/RDF knowledge graph. Batch enrichment job plus a FastAPI inference endpoint, fully containerized on Nebius.
Every number on the slide traces to its source.1
Internal PDFs become presentations where each figure resolves to an evidence ledger — an orchestrator fans out cite-only explorer agents per report, and a verifier gates the deck before it ships.
Kanban with an AI copilot
Multi-project boards with priorities, members, and an AI layer that summarizes columns and suggests next moves — the operations twin of how I actually run projects.
✦ summarize with aiMulti-service EV charging platform
Service-oriented architecture where every service owns its database, speaks sync and async, and degrades gracefully when neighbours fail. Built with a contract-first, three-phase agentic workflow.
Matter over Thread, custom devices as first-class citizens
Raspberry Pi 5 border router with a Nordic RCP radio, ESP32-C6/S3 nodes bridged into the Matter standard so home-built hardware behaves exactly like factory devices — same commissioning, same controls.
Agentic marketplace + MCP for market intelligence
Extended a market-intelligence platform with a custom agentic marketplace and an MCP layer — turning a dashboard product into something agents can operate natively.
Agentic sourcing for industrial procurement
Agent-driven vendor discovery and outreach, researched hands-on with truck, food, and solar manufacturers. Sunset in 2025 — the full post-mortem lives in the story below.
The education
three chapters, one question — dotted where the story continuesBSc Creative Technology
Came in a maker, left an experience researcher. AR classifiers, a gamified workout copilot, an installation that made the weight of child labour tangible — and a closing thesis that mapped process-mining visualizations to the information needs of each corporate tier. Mental languages, formalized.
Cloud Computing minor
A semester of treating infrastructure as code: provisioning AWS through Terraform until the cloud stopped being a place and became a material. Hover — the stack assembles the way a plan should.
MSc Business & IT — enterprise track
The integration layer: where AI systems meet business cases. TrenData's agentic marketplace, the decentralized digital twin, and VoltStream were all built here — around the question this whole site asks: what makes an organization actually adopt?
Pressure tests
rapid builds — each one a story and the skills it burned inAge-aware digital risk education
One insurance story, retold per generation: identity theft and impersonation phishing explained in each age group's own mental language — the rings widen, the message adapts.
On-premise LLM for medical claim review
A secure, on-prem inference platform assisting medical reviewers on complex claims — retrieval-augmented with prose detection, built in the season when most enterprises were still deciding whether to trust LLMs at all. The lock stays closed; the tokens still flow.
The geofence twin, under deadline
Built solo with the spec-driven agent framework and shipped ahead of the deadline — the same system that opens this portfolio, forged as a pressure test.
The toolbox
methods and stack — what I actually run teams, agents, and systems onSpec-driven agentic development
Strategist / implementer / operator roles, per-module guardrails, adversarial read-only verification, and a live run as the definition of done. How I ship with Claude Code without losing the plot.
Mental languages
Process Communication Model applied to teams: find the baseline each person thinks in, then explain in their language. Hover — two signals phase-locking is the whole method.
The card catalogue
Every teammate as a card: how they think, what energizes them, what language lands. Team formation becomes deck-building — and burnout becomes visible before it happens.
Motivation archaeology
From the business tracking school: founders rarely know what actually drives them. Dig past the strata of stated goals, find the real fuel — and lean iteration starts powering itself.
Energy before mood
Your state is secondary: no mood arrives on its own. Act with energy and relentlessness first — the effort changes the state, and the state follows the work.
Let yourself be bad
Detach the skill from the self and iteration becomes cheap. Being bad in public is the entry fee to being good — the scribble is how the clean line gets drawn.
The statusline
the cockpit I actually fly claude code with — steal itOff the clock
the human layer — also live, also reactsThe shelf
currently reading — hover the spinesYou are the architect of yourself.
Anything within the laws of nature is buildable — most people just never start. I'd rather build. If you're building too, let's find out what our systems can do together.