I build AI agents and GenAI systems that work in production.

Freelance AI engineer in Gothenburg, Sweden. I help startups and scaleups go from idea to deployed AI, with a focus on agentic architectures, LLM integrations, and production-grade infrastructure.

What I Build

AI Agents & GenAI Systems

I design and build agentic AI systems, from single-purpose LLM integrations to multi-step autonomous pipelines. This includes RAG systems, tool-using agents, structured output pipelines, and custom integrations with APIs and business systems. I work across the full stack: prompt engineering, evaluation frameworks, guardrails, and deployment.

Claude API
OpenAI API
MCP
Pydantic AI
Structured Outputs
Vector DBs

Production ML & MLOps

Not everything is an LLM. When the problem calls for custom models (computer vision, forecasting, classification) I build and deploy them with proper MLOps: automated training, model serving, monitoring, and CI/CD. A decade of experience shipping ML systems means fewer surprises in production.

Python
PyTorch
FastAPI
Docker
AWS / GCP / Azure

AI Strategy & Technical Advisory

I help CTOs and product teams figure out where AI actually fits, and where it doesn't. Build-vs-buy analysis, vendor evaluation, roadmap planning, and technical due diligence. Grounded in hands-on engineering, not slide decks.

Selected Work

Recent projects across fintech, trading, and automotive.

Fintech / Accounting

AI Transaction Categorization

Built an AI-powered bookkeeping integration that automatically categorizes financial transactions using Claude's structured output API. Designed for the Fortnox ecosystem (600K+ potential users). Includes a production MCP server wrapping the Fortnox API for seamless agent interaction.

Claude API
Structured Outputs
MCP
FastAPI
Docker
Financial Markets

Real-Time Trading Signals

Developed an algorithmic trading system for Swedish equities that processes real-time financial news via websocket, classifies signals using LLMs, and executes trades with a two-phase exit strategy. Includes a nightly LLM-based performance audit loop.

Python
LLM APIs
WebSockets
Real-time Data Processing
Automotive

Computer Vision — Gaze & Eye Tracking

Spent years building production computer vision systems for automotive gaze and eye tracking at a publicly traded company. Deep experience with annotation pipelines, model training at scale, and deploying ML in safety-critical environments.

PyTorch
Computer Vision
MLOps
Edge Deployment

How I Work

Step 01

Discovery

We talk about your problem, data, and goals. I'll tell you honestly if AI is the right tool.

Step 02

Proposal

A clear scope with timeline and cost. No ambiguity.

Step 03

Build

Iterative development. You see working demos early, not a big reveal at the end.

Step 04

Ship & Support

Production deployment, documentation, handoff. I stay on until it runs smoothly.

About

Based in Gothenburg, Sweden. Available worldwide.

Patrik Larsson — AI/ML engineer and founder of Tensorhead

Patrik Larsson

AI/ML Engineer

I'm Patrik Larsson, an AI/ML engineer with 10+ years of experience building production systems. I started in computer vision and classical ML, and now I spend most of my time building agentic AI systems and LLM integrations.

I run Tensorhead AB, a one-person consultancy based in Gothenburg. I work with startups and scaleups who need senior AI expertise without the overhead of a full team. Remote-first, available across Europe and beyond.

Gothenburg, Sweden
Tensorhead AB
AI Agents, GenAI, Production ML

Technologies

Python
Claude API
OpenAI API
MCP
AI Agents
RAG
Structured Outputs
Pydantic AI
Evals
Vector DBs
FastAPI
Docker
AWS / GCP / Azure

Let's Build Something

Tell me about your project and I'll get back to you within 24 hours.