Magnus Friberg · AI consultant · Stockholm
AI systems that survive contact with production.
I'm an independent AI consultant. I build AI agents and automation inside companies — on site, together with your team — so the systems keep working long after I've left. I've been shipping NLP and machine learning to production since 2018: at Spotify, at SeenThis, and at my own startup Saltfish, which exited via share sale in 2026.
Consulting · MANDGIE AB
AI agents your team ends up owning.
Through my company MANDGIE AB I take on a small number of hands-on engagements: building AI agents and LLM automation with client teams, on site, so that the evals, integration patterns and failure modes live with the people who'll run the system. Recent results:
- An affiliate network: an agent now does the manual data gathering for key account managers — saving each of them 4–5 hours a day. Read the case study.
- A dental care company: a company-wide business-insight agent lets anyone in the organization ask what's going on in the business — grounded in the company's own data.
Track record · 2017 — today
Eight years of shipping AI, not slides.
-
Independent AI consultant · MANDGIE AB, Stockholm
Advising and training enterprises on production AI, and building agent systems together with their teams. More →
-
Co-founder & CTO · Saltfish
AI platform that turned one product capture into interactive demos, videos and docs for B2B SaaS — used by customers like Fathom (50,000 users onboarded through the platform), Bokio and Magma Math. One of two engineers in a four-person company: I built the platform end to end — LLM generation pipelines across multiple providers, AI-narrated product tours, an MCP integration — and owned production. Raised a $730K pre-seed led by Antler. Exited via share sale in 2026.
-
Lead Data & AI Engineer · SeenThis
Built the data platform end to end — real-time ingestion from the edge at ~10,000 events per second, ~900 million events a day, into BigQuery — and the ML layer on top: models tuning the streaming algorithm and forecasting the impression volumes SeenThis priced its product on.
-
Senior Data Scientist · Spotify
Led the four-person team behind Spotify's internal intent model: transformer NLP, adopted from BERT within months of its 2018 release, clustering user text from support, email and social media into intents. It became company-wide infrastructure — a searchable corpus for every team, intent spikes as incident early-warning — and is still in production today.
-
Data Scientist · Capgemini
Customer-facing data science consulting in Nordic retail — GANT, Gina Tricot, Tiger of Sweden.
Before all that: a BSc in Statistics (Stockholm University) and a BSc in Economics (University of Gothenburg). Building neural networks since 2014, and competing on Kaggle for ten years.
Projects
Things I build on the side.
Minne
A free, open-source Mac app. Tap right-Option in any text field and the reply you would have written is already there — drawn from what you work on, in your words, on your machine.
minne.sh →Tarsan
AI agents for Sweden's property and construction industry — resolving tenant cases, coordinating trades, and retiring the manual work.
tarsan.ai →Writing
Notes from production.
-
What eight years of production NLP taught me about LLM agents
From BERT at Spotify in 2018 to agent systems in 2026 — what changed, and what stubbornly didn't.
-
How one AI agent gives key account managers 4–5 hours a day back
An anonymized case study: what the agent does, how it was built, and the lessons that transfer.
-
What does an AI consultant actually do — and when should you hire one?
The job behind the title, and an honest checklist for when bringing one in makes sense.
Contact
Working on something?
Tell me what you're trying to build — or automate. I read and answer every email myself.
magnus@mandgie.comOr find me on X: @mangefriberg