Senior Full-Stack Engineer (ML & Data) Stockholm - Hybrid

Senior Full-Stack Engineer (ML & Data) Own our prediction models from training pipeline to customer-facing feature. Strong applied ML — calibration, evals, leakage — plus full-stack TypeScript to ship it. We build AI agents that audit and optimize e-commerce advertising. Our platform syncs live data from Google Ads, Meta, Merchant Center, and Google Analytics, runs it through a rule engine, LLM-powered audit agents, and our own prediction models, and turns findings into concrete actions marketers can apply with one click. What sets us apart is measurement: we connect real profit data from our customers' e-commerce platforms to their ad accounts and build models that tell them what their marketing actually returns — not what ad platforms self-report. The ML in our product isn't decoration. Prediction models, calibration, and evaluation pipelines are core to what customers pay for. We're a small team shipping fast on a modern, strictly-typed stack. You'd report directly to the CTO, with real ownership: features you design end-to-end, and models you own from training data to the number a customer sees on screen. The role This is a hybrid role — roughly 60% product engineering, 40% ML & data science. You'll ship full-stack features in our TypeScript monorepo and own prediction models as production software: framing the problem, building training and evaluation pipelines, monitoring calibration drift, and wiring outputs into the product. We're not looking for a research scientist, or a pure web engineer. We want someone who treats a model the way a good engineer treats a service: tested, monitored, versioned, and honest about its failure modes. What you'll do Own prediction models end-to-end — models that predict product returns and net profit per order — training data pipelines, feature engineering, evaluation (ranking quality and calibration), recalibration strategies, and serving model output into the product. Build and maintain data pipelines — SQL feature builders on PostgreSQL shared by training and serving, versioned dataset exports, backfills, and data quality checks that keep training data honest (no leakage, no silent schema drift). Design and build features across our TypeScript monorepo — tRPC procedures and NestJS services on the backend, React on the frontend — so model insights become actions marketers can apply with one click. Work on the AI audit agent — prompt design, rule evaluation, structured-output evals, and plumbing that turns LLM output into reliable, deduplicated actions across multiple model providers. Extend integration workers — sync campaign, product, and analytics data from Google and Meta APIs via Cloud Pub/Sub pipelines. Evolve the data model in PostgreSQL/Prisma — keep migrations, multi-tenancy, and role-based access clean as the product grows. Own quality — write tests (Jest on TypeScript, pytest on Python), review PRs, and keep our CI/CD pipeline to Cloud Run fast and boring. Our stack Backend — TypeScript, Node, js, NestJS + Fastify, tRPC, Zod, Prisma, PostgreSQL, ML & data — Python, pandas, scikit-learn, XGBoost, PyTorch, ONNX, SQL on PostgreSQL, BigQuery, Calibration pipelines, Bayesian MMM (Google Meridian) Model serving — FastAPI on Cloud Run · Cloud Run Jobs · GPU training on GCP Batch Frontend — React · Vite · Tailwind · End-to-end types via tRPC Infrastructure — Google Cloud Run · Pub/Sub · Docker · Terraform · GitHub Actions · pnpm workspaces AI & integrations — Multiple LLM providers · Google Ads · Meta · Merchant Center · GA4 What we're looking for 5+ years building production software, with strong TypeScript on both server and client — or strong TypeScript on one side plus deep Python, with the willingness to close the gap fast. Hands-on applied ML: you've trained, evaluated, and shipped prediction models that real users depended on. You can explain the difference between good ranking and good calibration, know what data leakage looks like, and have debugged a model that was confidently wrong. Fluent Python for data work (pandas, scikit-learn or similar) and strong SQL — comfortable with large datasets in PostgreSQL or BigQuery. Solid relational database instincts — schema design, migrations, and query performance in PostgreSQL (Prisma is a plus). Experience with event-driven or queue-based architectures (Pub/Sub, SQS, Kafka, or similar) and their failure modes. A habit of testing: you write tests because they let you move faster, not because someone told you to. Pragmatic product sense — you ask why before how, and you'd rather ship a focused, well-measured model than a state-of-the-art one nobody can maintain. Nice to have Bayesian modeling or marketing mix modeling experience (priors, ROI calibration, health diagnostics). Hands-on work with LLM APIs in production — prompting, evals, structured output, cost control, switching providers. Experience serving models on GCP (Cloud Run, GCP Batch, Vertex AI) and monitoring for drift. Familiarity with Google Ads, Meta, or Merchant Center APIs, or ad-tech / e-commerce domain experience. NestJS, tRPC, or GCP (Cloud Run, Artifact Registry, Workload Identity) in a previous role. How we work Small team, short feedback loops, no ceremony for ceremony's sake. Strict TypeScript, code review on everything, CI that deploys to Cloud Run on merge. Models are held to the same standard as code: evaluated before they ship, monitored after. We optimize for shipping value weekly and keeping the codebase a place people enjoy working in. We're a fast-growing team building AI agents that handle the repetitive work — so companies can focus on vision, strategy, and innovation. Here, you'll own ideas from first sketch to finished product, work closely with the founders, and see your impact immediately. We celebrate success, learn from challenges, and believe true magic happens through collaboration. Want to shape the future of AI-driven marketing? Interviews are conducted on an ongoing basis, so please submit your applicat