COMPLEX PROCESSES. CLEAR SYSTEMS.
Marvin Kamp. Engineer and co-founder. I connect domain processes, modern software and accountable AI — from the first idea through to operations.
MK1 · explore projects, processes and engineering
Software development¹
Steel-sector ERP¹
Owner-built portfolio projects
Products, platforms and processes. Choose a project to explore the problem behind it — and the engineering decisions holding it together.
Product for AI-assisted social posts from a user's own travel photos, with a publicly verifiable landing page and application.
Auditable control plane for AI-assisted business processes with role and policy models, durable workflows, signed runner protocols, human gates, MFA, backups and a clearly separated demo mode.
Self-hosted creator workspace for planning, publishing and analyzing a creator's own short-form content and channels.
Local sports-data and decision-support platform with multi-source ingestion, normalization, ELO, odds time series, coverage controls and a human final decision.
Local companion for Paperless-ngx with case workflows, a scan studio and verified archive handoff. A sanitized architecture description explains the design without private documents or access details.
Co-founded logistics platform matching self-employed truck drivers without their own vehicle with freight companies; Marvin's technical role and the platform concept are publicly documented.
Version-controlled workflow suite for RoamTale with a live-tested caption automation, daily digest and clearly separated posting and push templates.
Owner-built recruitment portfolio with two visual directions, procedural three-dimensional MK1 lettering and an accessible project explorer. No external 3D models or runtime libraries.
Owner-built projects are labeled separately from synthetic work samples. Private operational data stays private.
Owned travel photos become platform-specific content drafts. RoamTale connects .NET, AI generation, persona voice and a guided product interface.
Public product page · 25 October 2026 · no account data
MARVIN KAMPI care why a process exists, which exceptions make it difficult and how a team can actually work with the solution.
My owner-confirmed experience connects long-term software development with 15 years in steel-sector ERP. Owner-built products and the publicly documented co-founding of CargoTrouper add product responsibility to that domain depth.
Understand terms, state, data and exceptions.
Deliver in inspectable steps and preserve a way back.
From C# and ERP to Python, modern web products and AI pipelines. Every area leads to a concrete project or an explicitly labeled work sample.
Decompose business processes, define approval boundaries and integrate AI output through traceable state.
Business logic, backend services and incremental legacy modernization. Understand and preserve existing behavior first.
From data models and authentication to usable interfaces, dashboards and complete workflows.
Connect commercial documents with material reality: units, tolerances, partial quantities and business exceptions.
Stable contracts, partner mappings, explicit error semantics and controlled recovery across system boundaries.
Combine sources, control duplicates, preserve provenance and make coverage and quotas visible.
Automate files, import paths and verification reproducibly. Keep outcomes, failures and provenance explicit.
Deploy applications behind TLS, isolate internal services and ship releases with health checks and a rollback path.
Account for failure modes, unknown outcomes and duplicate delivery. Verify compatibility and recovery explicitly.
Make technology usable for others through executable examples, traceable decisions and concrete operational guides.
Model scanning, visual review, OCR and archive confirmation as separate states; preserve originals and page provenance.
From a concrete problem to a usable product. Connect engineering decisions with the workflows of the people using it.
These labs are synthetic. They isolate concrete engineering decisions and can be inspected independently of private products.
Executable reference lab that decomposes a business process into automated, assisted and human-owned steps with risk classification, approval boundaries and replayable evidence.
TypeScript, Node.js, policy engine, JSONL audit, human-in-the-loopLAB 02 / SYNTHETICRead-only scanner for exported n8n workflows that flags possible inline secrets, unprotected webhooks, missing retry or timeout strategies and recovery gaps.
Python, n8n workflow JSON, static analysis, CILAB 03 / SYNTHETICDependency-free .NET CLI for focused breaking-change checks on published API contracts, including paths, operations, responses, schemas, required fields and types.
C#, .NET 10, OpenAPI 3, CI compatibility gateLAB 04 / SYNTHETICExecutable synthetic C# lab for incremental modernization of long-lived business systems: ingest, validate and compare legacy data behind a modern CLI/API boundary.
C#, .NET, CLI, fixed-width data, JSON, characterization testsIntegrate AI into real business processes: state, approvals, APIs and recovery.
Relevant evidence ↗C#/.NET, APIs, EDI and data flows for systems whose failures affect the business.
Relevant evidence ↗Build and operate products from backend services to usable interfaces.
Relevant evidence ↗Work connecting domain processes with software: AI-assisted automation, C#/.NET backends, APIs and B2B integration, and full-stack products. I enjoy bringing structure to established workflows and making a useful first step actually usable.
No. My foundation is business processes, long-lived software and integration. AI extends that work where output remains inspectable. Data models, failure handling, understandable interfaces and operations matter as much as the model.
An owner-built product connects interface, domain workflow, integrations and delivery. A synthetic lab isolates one specific pattern so it can be inspected quickly. Both are valuable; I deliberately distinguish their provenance and evidence scope.
Evidence pages link to published labs, OpenAPI files, mapping examples, runbooks and sanitized case studies. They do not expose private applications or employer code. Each reference explains what can be inspected publicly.
Start with a concrete challenge: which process costs your team time, where do failures occur, and which system boundaries matter? From there we can establish fit, onboarding needs and a useful first contribution.
Looking for someone who understands processes, connects systems and thinks a product through to operations? Let’s talk.