I build AI systems that run in production — often where the data can’t leave the building.
I’m Patrik Larsson — shipping ML since 2018. These days it’s mostly LLM systems: agent architecture, evals, and the MLOps glue underneath.
Tensorhead is my one-person consultancy in Gothenburg. I work with startups and scaleups that need senior AI engineering without hiring a department.
SERVICES
What I do
Call it a generative AI consultant, an LLM consultant or a senior engineer — the job is GenAI in production, and it has to hold up.
AI Agents & LLM Systems
Agent architectures, MCP server development, RAG pipelines, structured outputs, evaluation harnesses. From prototype to a system your team can actually operate.
Model Training & Fine-Tuning
Training and fine-tuning custom deep learning models in PyTorch — data pipelines, evals as regression tests, and the MLOps to keep models trained. For LLM systems, honest advice up front: retrieval, structured outputs and evals usually beat fine-tuning.
Applied Computer Vision
Real-time and safety-critical computer vision in production: data and annotation pipelines, large-scale model training, edge deployment.
Technical Advisory
Build-vs-buy analysis, architecture reviews, AI roadmaps, and technical due diligence for CTOs and product teams.
GROUNDED IN CODE, NOT SLIDE DECKS.
ABOUT
About Tensorhead
I’m Patrik Larsson, a freelance AI engineer and AI consultant based in Gothenburg, Sweden. My background runs from real-time computer vision through model training at scale to LLM systems and agents — all in production, since 2018. Tensorhead AB is the independent practice that grew out of that work.
The practice is remote-first, with clients across Sweden and Europe — on-site in Gothenburg or occasional travel by arrangement. I work in Swedish and English, and take contracts directly or through consultant brokers.
COMPANY FACTS
- COMPANY
- Tensorhead AB
- ORG.NR
- 559438-7853
- VAT
- SE559438785301
- LOCATION
- Stavhopparegatan 5A, 416 59 Göteborg, Sweden
- LANGUAGES
- Swedish & English
- MODEL
- Remote-first, on-site by arrangement
- CONTRACTS
- Direct or via brokers
- AVAILABILITY
- Booking from September 2026
FAQ
Common questions
Who is Tensorhead?
Tensorhead AB is the consulting company of Patrik Larsson, a freelance AI engineer and AI consultant in Gothenburg, Sweden. One person, shipping production machine learning since 2018, working with companies across Sweden and Europe. Clients get senior AI engineering without hiring a department — the same person who scopes the work builds it, hardens it and hands it over.
Are you available for new assignments?
Tensorhead AB is booking new engagements from September 2026 — I’m on parental leave until then. Intro calls and scoping conversations are welcome now, so a project can start the moment I’m back: describe what you’re building, and we’ll have a plan and a first milestone ready before the calendar opens.
What kinds of AI projects do you take on?
AI agents and LLM systems in production are the core: agent architectures, MCP servers, RAG, structured outputs, evaluation. I also take on model training and fine-tuning, applied computer vision, and technical advisory — as a senior developer embedded in your team, or as an advisor.
Do you work remotely, or only in Gothenburg?
Tensorhead AB is remote-first, based in Gothenburg, Sweden. Most work happens remotely with clients across Sweden and Europe; on-site in Gothenburg or occasional travel works by arrangement. Day to day that means working inside your repos and your infrastructure — not from a weekly status deck. I work in Swedish and English.
Do you train and fine-tune models, or only build on top of LLM APIs?
Both. At Tensorhead I train and fine-tune custom deep learning models in PyTorch — data preparation, training runs, evaluation harnesses, deployment. For LLM systems I’ll often advise against fine-tuning: better retrieval, structured outputs and a solid eval suite solve most problems at a fraction of the cost. When custom training is the right call, it’s work I’ve done for years.
Do you take contracts through brokers?
Yes. Tensorhead AB takes contracts directly or through consultant brokers — both are routine, and the paperwork is straightforward either way. The contracting entity is Tensorhead AB, org.nr 559438-7853, VAT-registered and based in Gothenburg, Sweden. If your procurement runs through a preferred broker, name it in the first email and we’ll take it from there.
What does an engagement cost?
Hourly or daily rate depending on scope and duration; fixed price is possible for well-scoped builds. As a rough market benchmark, senior AI consultants in Sweden typically run 1,200–2,500 kr per hour — large firms often charge more. Describe what you’re trying to do and you’ll get a concrete quote after the first call.
What is an MCP server — and have you built one?
MCP — the Model Context Protocol — is an open standard that lets AI assistants and agents call your systems and data through one well-defined interface. Yes: I have built and shipped a production MCP server, and building them for clients is part of the service — including the auth, rate limiting and evals that separate a demo from a real integration.
Can you build AI systems that are GDPR-compliant and keep data in the EU?
Yes — much of my work runs where the data can’t leave the building. In practice that means on-prem or EU-region deployments, data minimization, and evaluation suites that prove how the system behaves. I architect for GDPR from day one, and you get documentation your data-protection people can actually review.
CONTACT
Get in touch
Tell me what you’re working on.
Booking new engagements from September 2026 — intro calls welcome now. I reply within a day.