Back to site

Applied AI Engineer

Carsten Nilsson

Senior software engineer translating product and platform needs into reliable AI-enabled systems

Senior software engineer with long-running ownership of commercial software and hands-on leadership in practical AI adoption. I build production-minded AI workflows around Codex, MCP/tool integrations, provider routing, grounded context, validation, and test-backed engineering delivery.

Export CV

Export CV
  • 10+ repeatable AI workflows Built for triage, validation, documentation, operations, and backlog execution
  • 20x chart update speedup Shipped caching and batched shape deltas in Ampler Charts
  • DKK 10M+ ARR product context Senior ownership in mature commercial software
  • 1-2 product changes weekly AI-assisted TDD with recent PRs near 90% coverage
01

Engineering profile

Where I fit

I am strongest where ambiguous product, platform, or operator needs have to become working software. I stay close to the workflow, translate messy requirements into a technical path, ship the useful version, and keep feedback loops tight.

Practical AI work

My AI work is practical and hands-on: leading internal AI enablement at Ampler, teaching weekly tool workflows, using Codex daily, shaping context and validation habits, and turning AI output into tested, reviewable engineering changes.

Applied AI engineering

My current AI focus is making AI-assisted engineering reliable: connecting context, MCP/tool access, provider routing, validation prompts, and evaluation loops so plausible AI output becomes maintainable product change.

02

Experience

2018 — Now

Senior Software Engineer

Ampler

Mar 2018 - Present

Senior engineer owning core areas of a mature DKK 10M+ ARR commercial product, while leading practical AI adoption through weekly enablement, Codex workflows, evaluation habits, and test-backed delivery practices.

  • Carried Ampler Charts from growth-stage momentum into long-term maturity as a core commercial product.
  • Lead practical AI adoption at Ampler by teaching weekly tool workflows, shaping Codex usage patterns, and turning AI output into reviewable, tested engineering changes.

Built and evolved commercial productivity software across PowerPoint, Excel, Word, and Outlook, with primary focus on Ampler Charts. Worked hands-on in a mature C#/.NET desktop product where compatibility constraints, formatting fidelity, host behavior, and regression risk all mattered. The work centered on translating complex workflow and product needs into reliable implementation through profiling, targeted performance work, algorithmic improvements, safer exception handling, test-backed maintenance, and durable product design. I now lead practical AI adoption inside Ampler through weekly enablement and reviewable AI-assisted engineering habits.

  • Carried Ampler Charts from growth-stage momentum into long-term maturity as a core commercial product.
  • Lead practical AI adoption at Ampler by teaching weekly tool workflows, shaping Codex usage patterns, and turning AI output into reviewable, tested engineering changes.
  • Ship 1-2 product changes weekly using AI-assisted test-driven development; recent PRs reach near 90% coverage and 10-20 tests, up from 1-3.
  • Reduced PR regression stalls through stronger mocking, validation, and AI-assisted investigation.
  • Improved chart update performance by up to 20x by introducing caching and batching shape updates as deltas instead of applying changes one by one.
  • Implemented a research-based label placement algorithm for a hard overlap-reduction problem, keeping complex chart layouts workable in practice.

Tools: C#, .NET Framework, WPF / XAML, MVVM, PowerPoint automation, AI-assisted development, AI with test-driven development, Agentic coding workflows, AI context management, Performance optimization, Test automation, PowerShell automation, VSTO, Office Interop

More detail

Built and evolved commercial productivity software across PowerPoint, Excel, Word, and Outlook, with primary focus on Ampler Charts. Worked hands-on in a mature C#/.NET desktop product where compatibility constraints, formatting fidelity, host behavior, and regression risk all mattered. The work centered on translating complex workflow and product needs into reliable implementation through profiling, targeted performance work, algorithmic improvements, safer exception handling, test-backed maintenance, and durable product design. I now lead practical AI adoption inside Ampler through weekly enablement and reviewable AI-assisted engineering habits.

  • Ship 1-2 product changes weekly using AI-assisted test-driven development; recent PRs reach near 90% coverage and 10-20 tests, up from 1-3.
  • Reduced PR regression stalls through stronger mocking, validation, and AI-assisted investigation.
  • Improved chart update performance by up to 20x by introducing caching and batching shape updates as deltas instead of applying changes one by one.
  • Implemented a research-based label placement algorithm for a hard overlap-reduction problem, keeping complex chart layouts workable in practice.
  • PowerPoint automation
  • AI-assisted development
  • AI with test-driven development
  • Agentic coding workflows
  • AI context management
  • Performance optimization
  • Test automation
  • PowerShell automation
  • VSTO
  • Office Interop
  • C#
  • .NET Framework
  • WPF / XAML
  • MVVM

2014 — 2018

Senior Analyst

Accenture

Sep 2014 - Feb 2018

Delivered enterprise implementation work across ETL, migration, and business-critical backend systems where correctness, delivery discipline, and system constraints mattered.

  • Delivered ETL, data migration, and integration work where correctness mattered to client operations.
  • Worked across analysis, implementation, and stakeholder-facing delivery in structured enterprise environments.

Worked across analysis, implementation, and client delivery in enterprise consulting engagements. Contributed to .NET-based ETL pipelines moving business-critical data from IBM Db2 systems into a new database environment and to a ship maintenance and management system, combining hands-on implementation with the delivery discipline needed in constrained, risk-sensitive environments.

  • Delivered ETL, data migration, and integration work where correctness mattered to client operations.
  • Worked across analysis, implementation, and stakeholder-facing delivery in structured enterprise environments.
  • Handled business-critical backend workflows spanning legacy data sources, target-system constraints, and delivery expectations.

Tools: C#, .NET, ETL, IBM Db2, Data migration, Technical analysis, Client delivery, Enterprise systems

More detail

Worked across analysis, implementation, and client delivery in enterprise consulting engagements. Contributed to .NET-based ETL pipelines moving business-critical data from IBM Db2 systems into a new database environment and to a ship maintenance and management system, combining hands-on implementation with the delivery discipline needed in constrained, risk-sensitive environments.

  • Handled business-critical backend workflows spanning legacy data sources, target-system constraints, and delivery expectations.
  • Data migration
  • Technical analysis
  • Client delivery
  • Enterprise systems
  • C#
  • .NET
  • ETL
  • IBM Db2
Earlier roles 3 student software roles before Accenture

2014 — 2014

Student Software Developer

ClearView Trade

Feb 2014 - Aug 2014

Early commercial product development role held while finishing university studies.

  • C#
  • .NET
  • KnockoutJS
  • Commercial product work

2013 — 2014

Student Software Developer

Netcompany

Jun 2013 - Feb 2014

Early consulting-focused software role in a fast-paced team environment.

  • C#
  • .NET
  • ASP.NET
  • MSSQL

2012 — 2013

Student Software Developer

Infomedia

Dec 2012 - May 2013

First professional software role alongside university studies.

  • C#
  • .NET MVC
  • MySQL
  • Professional delivery
03

AI Systems & Agentic Tooling

I build repeatable AI engineering workflows: custom Codex skills, repo-aware agents, MCP/tool integrations, prompt workflows, validation suites, and automation loops that turn AI assistance into reviewable engineering output.

  • Built 10+ repeatable AI workflows for triage, validation, documentation, operational review, and backlog execution.
  • Designed an AI operations platform around Open WebUI, Codex-LB, MCP, provider routing, curated context, and validation suites.
  • Created agent guardrails for infrastructure work: controller-first validation, rollback-aware change windows, read-only discovery, and mutation boundaries.
  • Maintained AI-assisted workflows that connect GitHub issues, local repo docs, operational inventory, runbooks, and automated validation.

AI Tooling

Codex, Open WebUI, Codex-LB, MCP, prompt packs, agent workflows

Automation

GitHub CLI, Ansible, OpenTofu/Terraform, validation scripts

Platform Ops

Proxmox, Cloudflare, Caddy, Docker, monitoring

04

Selected systems

AI platform

AI platform operations and agentic workflow systems

Operate and evolve a private AI workspace with 10+ repeatable workflows for investigation, validation, documentation, operational review, and backlog execution across Open WebUI, codex-lb, provider routing, curated knowledge, MCP/tool integrations, and evaluation prompts.

  • Open WebUI
  • codex-lb
  • OpenAI-compatible APIs
  • Model and provider routing
  • MCP / tool calling
  • Knowledge packs
  • AI evaluation loops
  • AI context management

Keywords: Open WebUI, codex-lb, OpenAI-compatible APIs, Model and provider routing, MCP / tool calling, Knowledge packs, AI evaluation loops, AI context management

  • Use Codex daily across multiple worktrees, setting up long-running agentic sessions and subagent orchestration with clear context, guardrails, and review checkpoints.
  • Built repo-managed AI context flows around curated documentation, knowledge packs, validation prompts, and read-only tool access.
More details
  • Maintain an OpenAI-compatible provider path through codex-lb and Open WebUI so model/provider behavior can be routed, inspected, and improved.

Agentic workflows

  • Structure work so AI agents can run for hours with enough context to make progress and enough boundaries to keep the result reviewable.
  • Use 10+ workflows for app onboarding, monitoring, drift checks, network state, bottlenecks, CPU caps, alerts, and recovery.
  • Use parallel worktrees to explore, implement, and compare changes without losing control of repo state.

Platform and quality

  • Operate Open WebUI with curated workspaces, hosted and local provider views, Tavily-backed research, and repo-first knowledge behavior.
  • Maintain codex-lb as an OpenAI-compatible backend/provider layer with private API routing and protected dashboard access.
  • Use validation prompts and knowledge-quality checks to catch generic, stale, or ungrounded AI answers before treating them as useful.

Private platform

Private platform operations and infrastructure automation

Operate a private platform used to test production-style AI and infrastructure workflows, including automation, observability, access control, AI runtime support, and repeatable day-2 operations.

  • Proxmox
  • Ansible
  • Terraform / OpenTofu
  • Docker / Docker Compose
  • Cloudflare Access
  • Caddy
  • Tailscale
  • Prometheus / Grafana / Alertmanager
  • Linux systems administration
  • GitHub Actions / CI automation
  • AI platform operations

Keywords: Proxmox, Ansible, Terraform / OpenTofu, Docker / Docker Compose, Cloudflare Access, Caddy, Tailscale, Prometheus / Grafana / Alertmanager, Linux systems administration, GitHub Actions / CI automation, AI platform operations

  • Own the stack from virtualization and container runtime to ingress, access policy, AI workspace support, monitoring, verification, and day-2 operations.
  • Run the platform through Ansible, Terraform/OpenTofu, and documented workflows instead of ad-hoc administration.
More details
  • Use inventory, validation, drift checks, and post-apply verification as standard operating practice.

Platform

  • Built around Proxmox VMs and LXCs with documented inventory, lifecycle management, and sizing decisions.
  • Operate Linux-based services through VMs, LXCs, and Docker Compose with repeatable storage and backup workflows.
  • Support the platform with local tooling, AI-assisted workflows, and automation so recurring operations stay reviewable and repeatable.

Access and operations

  • Manage infrastructure through Ansible, OpenTofu, and Terraform, including Cloudflare DNS, Tunnel, and Access workflows.
  • Built ingress and reachability patterns with Caddy, Tailscale, and private DNS across public and private access tiers.
  • Expanded observability with Prometheus, Grafana, Alertmanager, blackbox checks, service-specific monitoring, and AI-path validation.

Developer tooling

Cross-platform Go CLI and terminal tooling

Built a cross-platform Go CLI and terminal-first operator tool for Codex usage, account switching, isolated runtimes, local state, diagnostics, support capture, and feedback loops.

  • Go
  • CLI / developer tooling
  • Terminal UI / TUI
  • SQLite / local persistence
  • Debugging and diagnostics tooling
  • Prometheus / observability instrumentation

Keywords: Go, CLI / developer tooling, Terminal UI / TUI, SQLite / local persistence, Debugging and diagnostics tooling, Prometheus / observability instrumentation

  • Designed an operator tool for managing Codex auth profiles, account switching, and isolated runtimes rather than a thin wrapper.
  • Combined CLI flows, terminal UI state, local persistence, diagnostics capture, and lightweight metrics in one maintainable tool surface.
More details
  • Shipped with testing, mock scenarios, cross-platform CI, and contributor-facing documentation.

Build

  • Built in Go with a structured command surface for Codex usage, account/profile switching, terminal-first runtime views, history handling, and local state management.
  • Used SQLite-backed local persistence for analytics and history instead of treating tool state as disposable.
  • Designed around real operator workflows: quick diagnosis, repeatable action, and enough state to support later review.

Operate

  • Added debug bundle and snapshot workflows to make support and first-step diagnosis easier.
  • Exposed lightweight HTTP and Prometheus-facing metrics so the tool can participate in observability workflows without turning into a backend service.
  • Kept feedback loops short by making local behavior inspectable rather than hidden inside one-off commands.
05

Skills

AI Tooling

AI-assisted development, Agentic coding workflows, AI context management, AI evaluation loops, Open WebUI, OpenAI-compatible APIs

Core Engineering

C#, .NET / .NET Framework, Go, TypeScript, Python, SQL

Quality & Runtime

Performance optimization, Profiling, Reliability / exception handling, Debugging and diagnostics tooling, Testing

Automation

CI/CD, Docker / Docker Compose, GitHub Actions / CI automation, Terraform / OpenTofu, Ansible, Infrastructure as code

Platform Ops

Platform operations, Linux systems administration, Observability, Prometheus / Grafana / Alertmanager, Access control

Networking

Cloudflare Access / Tunnel, Tailscale / private networking, Private DNS and ingress, Caddy

Specialized

Microsoft Office add-ins, PowerPoint automation, VSTO, Office Interop, ETL

AI Tooling

  • AI-assisted development
  • Agentic coding workflows
  • AI context management
  • AI evaluation loops
  • Open WebUI
  • OpenAI-compatible APIs
More skills
  • MCP / tool calling
  • Model and provider routing

Core Engineering

  • C#
  • .NET / .NET Framework
  • Go
  • TypeScript
  • Python
  • SQL
More skills
  • CLI / developer tooling

Quality & Runtime

  • Performance optimization
  • Profiling
  • Reliability / exception handling
  • Debugging and diagnostics tooling
  • Testing

Automation

  • CI/CD
  • Docker / Docker Compose
  • GitHub Actions / CI automation
  • Terraform / OpenTofu
  • Ansible
  • Infrastructure as code
More skills
  • Test automation

Platform Ops

  • Platform operations
  • Linux systems administration
  • Observability
  • Prometheus / Grafana / Alertmanager
  • Access control

Networking

  • Cloudflare Access / Tunnel
  • Tailscale / private networking
  • Private DNS and ingress
  • Caddy

Specialized

  • Microsoft Office add-ins
  • PowerPoint automation
  • VSTO
  • Office Interop
  • ETL
More skills
  • WPF / XAML
  • MVVM
06

Education

Danish grades shown in the original scale with an approximate US equivalent.

M.Sc. in Computer Science

2012 - 2014

Technical University of Denmark

Graduate studies in computer science and software engineering, with focus on architecture, requirements, formal methods, and a deeper technical foundation for professional software engineering.

Grade: 9.5 | US: A-/B+

Thesis Development of Social Network for Local Communities using a Cloud Platform + more details - fewer details

Designed and built an Azure-based social platform for local communities to share resources and household items.

  • Built around an Azure/cloud platform with a local-community sharing use case.
  • Focused on practical resource sharing and household-item reuse.
  • Supervised by Christian D. Jensen.

Grade: 10 | US: A/A-

Thesis: Development of Social Network for Local Communities using a Cloud Platform

Designed and built an Azure-based social platform for local communities to share resources and household items.

  • Built around an Azure/cloud platform with a local-community sharing use case.
  • Focused on practical resource sharing and household-item reuse.

Thesis grade: 10 | US: A/A-

  • Recognition: DTU Blue Dot Award (2013) for significant extracurricular work in DTU sustainability engineering projects.
  • Specialized in software engineering with emphasis on architecture, requirement elicitation, and mathematical formalization.
  • Built on earlier engineering work with more advanced software and computer science study.
  • Completed the degree while already gaining hands-on industry experience in student developer roles.

B.Sc. in IT & Communication Technology

2009 - 2012

Technical University of Denmark

Broad engineering foundation across mathematics, natural science, networks, software development, and communication technology.

Grade: 7.4 | US: B

Bachelor project Building an Intelligent Controllable Home - For the Solar Decathlon House + more details - fewer details

Built an iPad-based control system for an energy-positive competition house, covering appliance control and house telemetry.

  • Built for the Solar Decathlon energy-positive house context.
  • Covered both appliance control and house telemetry.
  • Supervised by Christian D. Jensen.

Grade: 10 | US: A/A-

Bachelor project: Building an Intelligent Controllable Home - For the Solar Decathlon House

Built an iPad-based control system for an energy-positive competition house, covering appliance control and house telemetry.

  • Built for the Solar Decathlon energy-positive house context.
  • Covered both appliance control and house telemetry.

Bachelor project grade: 10 | US: A/A-

  • Established the base for later work in software development, problem solving, and technical delivery.
07

Engineering signal

Activity snapshot

Loading anonymized activity signal.

Full activity overview

Loading rhythm.

Loading scope.

Recent activity mix

Recent activity