Alex Batten, AI Engineer in Denmark
Hi, I'm Alex, an AI engineer in Denmark with a BEng in Software
Technology from DTU. My background is C# and .NET: backend services,
REST APIs, and enterprise integrations.
Available for new freelance and contract work: AI agents and AI features,
automation, AI guardrails and evaluation, enterprise and tax-authority
integrations (SKAT, SAF-T, NemHandel/Peppol), .NET and full-stack web
applications, REST APIs, and Stripe billing, remote across the EU on CET. Email
alex.batten1234@gmail.com
or send a brief through the form on this site. The
full CV also answers
frequently asked questions about availability, stack,
and how to start a project.
Experience
Senior Software Engineer, AI Infrastructure at Alignerr (Contract, signed 2026, not yet started)
Signed contract, first assignment still pending. The agreed scope is
the infrastructure behind AI training, evaluation, and deployment
pipelines, in Python and TypeScript. Stack: Python, TypeScript.
AI Engineer & Co-Owner at Acconta (2026–Present)
Building a multi-tenant Danish bookkeeping SaaS across the stack: microservices
in ASP.NET Core and EF Core on .NET Aspire and Azure, with OpenAPI/NSwag generated
typed clients, and the Next.js client on top. Moved six services onto a shared
service base for startup, configuration, health checks, OpenAPI docs, and
authorization. Took VAT reporting to SKAT from the 17-field return calculation
through to the Next.js screens and SKAT's production gateway. The integration signs
its NemVirksomhed (RSU) SOAP calls with WS-Security, using a certificate held in
Azure Key Vault. Built the bank reconciliation backend on an open banking feed
(Aiia): matches that must sum exactly to the bank line, a reconciliation overview,
and an idempotent transaction sync in its own service. Upgraded SAF-T import and
export to the 2.1 schema, with tenant isolation and explicit refusal rules, ahead of
the 2027 audit-file mandate. Owning Stripe billing end to end: subscriptions and
seat logic (proration, credits, plan switching, yearly plans, plan lapses),
webhooks, and secret hardening. Built integrations for daily currency rates
(Danmarks Nationalbank), CVR company data (Danish Business Authority), and
transactional email. Leading e-invoicing over NemHandel and Peppol (OIOUBL, Peppol
BIS) and building the service that sends and receives invoices. Wrote the
requirements spec, ran the access point provider selection, and clarified the
registration model with the Danish Business Authority. Turning the Bookkeeping Act
and its executive orders into implementation specs and reviewing the platform's
authentication and multi-tenant security. Doing most of the team's in-depth code
review. Set up an autonomous AI development agent (an OpenAI dot) that the team
assigns GitHub issues to, and developed its operating prompt from a colleague's
first draft over seven revisions. The agent is built to hand each change to Codex
agents that implement, review, and test it, then open a pull request for human
review. Wrote and presented the team's guide to AI-assisted development with Claude
Code. Stack: C#, .NET Aspire, ASP.NET Core, Entity Framework, Microservices, Azure,
Stripe, Next.js, NemHandel, Peppol, AI Agents.
AI Safety Expert at Mercor (Contract, 2026–Present)
Writing prompts and conversations that test how AI models handle harmful
and sensitive requests, and pointing out where their responses fall short of
safety guidelines. The finished work becomes safety training data for large
language models. Skills: AI Safety, Red Teaming, Prompt Writing, LLM Training
Data.
Generalist Expert, AI Evaluation at Mercor (Contract, 2026–Present)
Evaluating AI model outputs against structured evaluation guidelines and
writing the rationale behind every judgment. The work is close reading:
checking a response for factual accuracy, sound reasoning, instruction
following, and tone, then stating precisely where it holds up and where it
breaks down, with evidence. Applying the same standard consistently across a
wide range of subject areas so that assessments stay comparable between
reviewers. Skills: AI Evaluation, LLM Outputs, Structured Feedback, Technical
Writing.
Generalist Danish Annotator at Mercor (Contract, 2026)
Wrote prompts in natural Danish on topics of general interest and annotated
how AI models responded to them. The annotations pointed out errors in the
responses, including errors in their Danish. The finished work became
training data for large language models. Skills: Data Annotation, Danish
Language, Prompt Writing, LLM Training Data.
AI Data Annotator at Outlier (2025–Present)
Contributed to advanced AI model development through data annotation,
prompt engineering, trajectory refinement, and QA across projects including
Clutch Zayu, Maraca Camera, Mechanic Astrologer, and Meter Pavilion
(OpenClaw). Crafted system prompts defining model behavior and constraints,
and designed single and multi-turn conversational and agentic datasets
spanning 50+ risk categories including bias, misinformation, privacy, and
safety. Built autonomous agent trajectories exercising skill discovery,
memory, long-horizon context, and native tools (cron, sub-agent delegation).
Served as both author and reviewer on Mechanic Astrologer. Skills: AI/ML, Prompt
Engineering, Agentic AI, Data Annotation.
Software Development Intern at IT Operators / Simple Agency Group (2024)
Developed REST APIs with auth, plus usage-limit monitoring with email
alerts. Built fullstack web apps with Angular, .NET, and Bootstrap,
including JWT auth, real-time dashboards, and automated reporting.
Migrated an in-house ERP's time representation from quarter hours to
minutes across every display, estimate and report field, then rebuilt the
billing review on top of it so a task worked by employees from more than
one group billed under each of those groups instead of one. Validated the
rebuilt invoicing screen by reconciling a full month of billing data
against what had actually been invoiced.
Projects
Custom REST API for Uno X (Client: ISTOBAL)
Custom API for ISTOBAL (car wash manufacturer) serving operational data to
Uno X (Nordic fuel station chain), with token-based authentication and
per-call usage counting with automated email alerts on a usage limit.
Automated SMS alert system (Client: Eye4u)
Extended a customer monitoring portal with automated SMS notifications
triggered by medicine cooler temperature deviations across a sensor network.
Added an adjustable pre-alarm tier that warns by SMS and email without
raising an incident, with suppression logic so a sensor sitting outside its
threshold does not re-notify on every check. Built as a multi-tenant module,
each customer with their own sensor configuration and alert thresholds.
Internal portal (Client: Dexa Bodyscan)
Built the initial web application for a body-scan clinic's internal
portal, from an empty repository to a working authenticated application in
eight working days: project scaffold, Dockerfile, DNS record, SSL
certificate, MySQL schema, Entity Framework contexts, JWT login, salted
password hashing, role-gated pages, and password reset from an email
link.
Internal Monitoring Dashboards (IT Operators)
Real-time dashboards deployed on Raspberry Pis for office TV screens,
monitoring time registrations and cron job health.
External API integrations, TIMEVAT
Integrated the EU VIES API for VAT number validation and currency exchange
rates from the European Central Bank into a VAT compliance platform.
Accelo to Business Central integration (Bachelor Project, DTU, 2025)
Fullstack application, made in collaboration with IT Operators / Simple
Agency Group, that imports time registrations from Accelo (PSA) into MSSQL for
review: the Accelo side of an integration with Microsoft Business Central
(ERP). The Business Central side was designed against its API v2.0 but not
built. Custom review UI, MSSQL data model, single sign-on via Microsoft Entra
ID (single-tenant), Docker in a Jenkins CI/CD pipeline with SQL Server outside
the containers, unit tests on both layers, and end-to-end tests
(WebApplicationFactory, headless Selenium) for the core fetch and overview
flows.
Invarix Guard
A plug-and-play AI safety library for .NET, shipped as a NuGet package
(Invarix.Guard). Covers prompt injection, PII, and content safety so apps can
add guardrails in a few lines of code.
Website: invarix.dk.
NuGet: nuget.org/packages/Invarix.Guard.
Invarix Guard Evidence
A free .NET library (Invarix.Guard.Evidence), licensed under Elastic License 2.0,
that records AI decisions as an append-only audit trail. Each decision is stored as a
CloudEvents event keyed on SHA-256 hashes rather than the prompt text.
Batches are sealed into an RFC 6962 Merkle tree and signed with the customer's
own Ed25519 key, so an auditor can prove no record changed after sealing using
standard Certificate Transparency tooling. Includes retention with legal holds,
signed deletion certificates, and a streaming export endpoint. Runs fully in
process with no outbound calls.
Website: invarix.dk.
NuGet: nuget.org/packages/Invarix.Guard.Evidence.
Invarix Gate
A deterministic action firewall for AI agents, shipped as a NuGet package
(Invarix.Gate). It sits in front of an agent's tool calls and evaluates each one
against a policy before it runs, returning allow, warn, escalate, deny, or
terminate. Eleven detectors cover destructive SQL, shell commands, git history
rewrites, cloud teardown, credential egress, runaway loops, protected targets, mass
sends, tools outside an agent's allowlist, call and token ceilings, and edits to
Gate's own policy. Ten of them distill real incidents, most with public
postmortems. Matching is pure CPU pattern work with no model call, so a verdict
lands in a median of 57 µs and is deterministic under a given policy. Policy is
authored in YAML or fluent C#, attaches through Microsoft.Extensions.AI and
Microsoft Agent Framework, and reports the surfaces it cannot see rather than
rendering a clean run. Ships with a test corpus of about 220 scenarios, more than
90 of which exist to prove ordinary work stays quiet.
Website: invarix.dk.
NuGet: nuget.org/packages/Invarix.Gate.
Hemi-Sync Guided Meditation App (paused)
A concept app for generating customizable guided meditations using
AI-driven text-to-speech layered with Hemi-Sync binaural beats, built in
React Native for iOS and Android.
Distortion VST Plugin
An audio distortion plugin with bitcrushing built with the JUCE framework
(C++), compatible with most DAWs as a VST.
Lab: writing and notes
nothing here yet.
Skills & Tools
Languages: C#, C++, Java, TypeScript, JavaScript, Kotlin, SQL.
Frameworks & Libraries: ASP.NET Core, Entity Framework,
Angular, .NET, .NET Aspire, Next.js, React Native, Bootstrap, JUCE.
Concepts & Practices: REST APIs, System Design,
Distributed Systems, Agile/Scrum, CI/CD, JWT Auth, Docker.
Other: Git, Claude Code, ERP Systems, Multi-tenant SaaS, SAF-T,
Audio Engineering.
Education
BEng Software Technology, Technical University of Denmark (DTU), 2021–2025
Electives in Computer Graphics, Rendering, C++ Programming, and Engineering
Economics. Bachelor's project: a PSA-to-ERP integration (Accelo to Microsoft
Business Central), with the Accelo side built from scratch and the Business
Central side designed but not built. GPA: 7.62 (Danish 7-point scale).
Recommendations
Mike Valencia, CEO at Simple Agency Group
Wrote about Alex Batten’s work at Simple Agency Group.
View the post on LinkedIn.
Erik Seifert
Endorsed Alex Batten for C#, .NET Framework, and Entity Framework (EF) Core.
Freelance & Project Requests
Available for new freelance and contract work. Email
alex.batten1234@gmail.com
or send a brief through the request form on this site. The work I take is
AI agents and AI features, automation, AI guardrails and evaluation,
enterprise and tax-authority integrations (SKAT, SAF-T, NemHandel/Peppol),
.NET and full-stack web applications, REST APIs, and Stripe billing, remote
across the EU on CET. Project requests get a reply within a couple of days.