Per WinrothProduct builder, AI systems. Malmö, Sweden.

    I spent ten years at Warner Music Group building products for the music industry. The biggest was Warner Connect, a marketing platform used every month by about 2,000 colleagues in 60 countries. Its campaigns reached more than 100 million fans and brought in over SEK 250 million. In 2026 I started building AI products on my own. Subfeel, Got Cosy and Got Hush are live.

    Work Experience

    I've spent my career building innovative products across the music and digital space.

    Independent Product & AI Developer

    Self-employed

    2026–present

    Building AI-native products end to end, solo.

    Parasition

    Advisory

    Sep 2025–Apr 2026

    Advised Parasition, whose AI Hugo is a marketing co-working space for the music industry, on product and go-to-market.

    Parental leave

    2024–2025

    Warner Music Group

    2014–2024

    Product Manager

    2023–2024

    Led WMG's direct-to-fan FanApp with a focus on privacy compliance and user experience.

    Product Director

    2015–2023

    Built Warner Connect from zero into WMG's most-used marketing tool. Campaigns reached 100M+ fans, 15M signed up, over SEK 250M in revenue. Led the global operational rollout of Topsify across about 60 markets.

    Current Projects

    Three live products shipped solo in 2026, and one on the way.

    Subfeel homepage

    Subfeel

    Visit

    The feeling underneath. Reads nine specific feelings in what people already wrote, with founder-judged ground truth and a locally fine-tuned model. Formerly Feel Layer; the measurement layer under Got Cosy and Got Hush.

    InstrumentMCPFeelings
    Got Cosy homepage

    Got Cosy

    Visit

    6,300+ hotels scored for cosiness from guest reviews. AI-consumable graph exposed via MCP.

    AI ScoringMCPHotels
    Got Hush? homepage

    Got Hush?

    Visit

    Launched July 2026. 813 hotels with evidence-gated quiet scores deduced from guest reviews.

    AI ScoringEvidence-GatedHotels
    2026
    Buy-Again Score
    buyagainscore.com

    Buy-Again Score

    Visit

    Product registers answering one question: would owners buy it again. Launching soon.

    Product RegisterReviewsLaunching Soon

    Core Skills

    My professional toolkit built across multiple industries and roles.

    AI systems and agents

    I run projects on DIE-OS, an operating system for AI agents I built and published: 14 constrained roles, 14 gates, append-only laws, a mechanical checker. I use frontier models for judgment, cheaper models for execution.

    Measurement and evaluation

    I judge nine feeling detectors blind against my own verdicts, run sealed exams with frozen bars and look budgets, and use an adversarial reviewer that files predictions before results exist. I made over 2,900 blind verdicts in the last month alone.

    Local models and fine-tuning

    I fine-tuned 14B models (QLoRA on rented L40S, MLX on Apple silicon) that beat frontier models on domain labeling. I moved image analysis to a free open-weight model on local hardware.

    Analytics and data

    I built the scoring pipelines behind Got Cosy (6,300+ hotels on a calibrated 0 to 10 scale) and Got Hush (813 hotels, evidence-gated), plus a corpus engine that discovers, licenses and ingests review datasets continuously.

    Product management

    I have twelve years of product management experience: I built Warner Connect from zero to WMG's most-used marketing tool, campaigns reaching 100M+ fans, SEK 250M+ revenue; now I ship solo, end to end, from data to live product.

    Stakeholder management

    I worked with global teams at Warner Music, artists and their management (Coldplay, David Guetta, Ed Sheeran), and executive stakeholders, and advised an AI marketing startup.

    CRM and marketing

    I built audience segmentation with predictive models that lifted email conversion from 12.5% to over 35%; I apply answer-engine optimization so AI assistants cite the products.

    Privacy and compliance

    I owned privacy for Warner Connect across campaigns reaching 100M+ fans and 10,000 digital campaigns, and privacy compliance for WMG's D2C FanApp across global territories.

    Want help with something like this?

    Email me.