AI Engineer building retrieval systems, agent orchestration, user-facing AI products and reliable execution infrastructure.

I combine applied AI with more than 20 years in enterprise IT, automation, operations and cybersecurity. My work focuses on systems that are useful in practice: searchable knowledge, controlled agent actions, product-facing AI workflows, auditability and operational safeguards.

Start with the live portfolio landing page
Recruiter proof packages, method profiles, supporting evidence and earlier showcases in one public route.

These four projects provide complementary evidence across retrieval, product engineering, agent interaction and runtime reliability.

source material and operational signals
-> structured ingestion and typed contracts
-> retrieval, policy or decision logic
-> user-facing React / web / desktop surfaces
-> explicit approval, evidence and audit boundaries
-> reproducible verification receipts

• Applied AI: hybrid retrieval, provider integrations, agent workflows and AI-assisted product features.
• Modern application engineering: Python/FastAPI, React/TypeScript, Astro, Electron and Supabase.
• Reliable execution: policy gates, reversible changes, deterministic decisions and durable evidence.
• Security-aware design: credentials kept out of browser or renderer surfaces, path confinement, workspace boundaries and explicit non-claims.

The detailed source pins, verification limits and publication status are recorded in docs/recruiter-proof-evidence-2026-08-02.md.

A scene-driven React and Three.js engine that normalizes agents, projects, workflow runs and graph events into a shared node/edge/event model.

The public-safe demonstration shows synthetic agents, focus, replay, priority filtering, provenance and snapshot export without exposing real messages, workflow data or private source.

Explore the Vertex visual proof

A Python CLI and SQLite action-evidence ledger that separates observed evidence, declared project context and derived reports.

The synthetic proof demonstrates temporal honesty, privacy receipts, secret-shaped-input refusal and JSONL/Markdown projections without exposing a real ledger, shell history, transcript or Personal RAG record.

Explore the Worktrace provenance proof

The source pins and verification boundaries for both supporting cases are recorded in docs/supporting-proof-evidence-2026-08-02.md.

A RunContext-governed orchestration runtime and multi-agent backbone. The current evidence route distinguishes the adopted compiled execution path, exposed through CLI and HTTP, from peer and reference surfaces such as AgenticLoop, OODA, scenario execution and review.

Open the current source-pinned SPINE proof · Browse the full showcase

A private teaching and reference implementation of capability-based routing, provider fallback, observability and optional learning with FastAPI, LanceDB and Neo4j. It is not presented as an active production service or as the estate's current universal MCP router.

Open the current source-pinned evidence route · Browse the historical showcase

A public, default-deny execution boundary for agent workflows with policy validation, environment scrubbing, resource limits, output caps and local-model support.

It is a bounded process-execution layer, not a kernel or container isolation boundary. The public source remains canonical; the evidence card provides a short recruiter orientation.

Read the source-first evidence card · Inspect the public source

The freshness pins, lifecycle boundaries and non-claims for these three cases are recorded in docs/supporting-proof-evidence-p1-02-2026-08-02.md.

A set of smaller public products demonstrates narrow, explainable MCP capabilities:

A controlled audio-first pipeline for a local morning brief in Västra Götaland:

approved public sources
-> collection and deduplication
-> ranking and rundown
-> Swedish script generation
-> editorial QA
-> voice and script-anchored captions
-> human review

The prototype is AI-assisted and human-reviewed. It is not presented as an autonomous newsroom or as permission to reproduce copyrighted source articles.

I prefer systems where claims can be inspected:

• source and implementation commits are pinned;
• tests and runtime receipts are distinguished from documentation claims;
• private data is replaced by synthetic fixtures in public demonstrations;
• limitations and non-claims sit beside strengths;
• risky actions pass through explicit policy or human approval;
• preserve-first change practices keep earlier evidence available.

This is also why many of the public proof packages are explanatory or synthetic. They demonstrate the product and control model without publishing credentials, prompts, private patches, operational logs or customer-like data.

Adaptivearts.ai is my independent applied-AI initiative for research, prototypes, technical writing and product experiments.

• Research and prototypes
• Articles and research
• MCP directory
• Context engineering
• AI-agent architecture

More than 20 years across enterprise IT operations, systems engineering, automation, cybersecurity and DevOps, including SOC/XDR work, identity and endpoint environments, monitoring, incident-oriented operations and business-facing technical delivery.

My current focus is the intersection of:

• applied AI and AI enablement;
• automation and integration;
• retrieval and knowledge systems;
• agentic workflows;
• security, governance and observability;
• translating technical capability into practical business value.

Occasionally the human also emits audio.

🎧 Spotify artist profile

The previous, broader profile README is preserved at archived/README.pre-canonical-recruiter-landing-2026-08-02.md.

_{Applied AI · Automation · Reliable agent systems · Adaptivearts.ai}