The Role
Among other things, you'll design and own the systems that decide which licensed data reaches our customers' models, and the evaluation infrastructure that proves it's better than the alternatives.
In this role, you will:
Design and operate the retrieval pipeline that serves licensed data to AI agents and models in real time: hybrid search, ranking, fusion, re-ranking
Build the eval harness that runs continuously, gates deploys on quality, and surfaces regressions before customers feel them
Translate research from our data science team into production systems, then own them end to end
Define what "good" means for Redpine's output and build the dashboards that prove it across customers, verticals, and competitors
Work across the stack as quality demands it, from ingestion signals through embedding pipelines to the API response
The Team
We're a small, fast growing team based in Stockholm, working across infrastructure, data, and AI. The same people designing systems are building, shipping, and running them. Feedback loops are short and ownership is direct.
You'll work closely with our data scientists, who own research, methodology, and data quality. You'll own production retrieval and the eval infrastructure that runs on top of that.
We're early enough that many of the core systems are still being defined. That means you won't be scaling someone else's architecture.
What we're looking for
Production experience with retrieval systems: search, ranking, recommendation systems, or RAG at scale, not just notebooks
Strong opinions on evaluation, both offline and online. You've designed test sets, run A/B experiments, and argued with data scientists and product managers about whether a metric reflects reality
Comfortable with the full retrieval stack: embeddings, vector and lexical indexes, learning-to-rank, hybrid systems & knowledge graphs
Have shipped, not only researched. You know the difference between a method that works in a paper and one that works in production
Comfort operating where direction evolves quickly and problems aren't yet well-defined
We work primarily in Go, Rust and Python in a multi cloud environment. We hire for judgment, not stack.
Why this role
Most retrieval and AI engineering roles are about plugging the latest models into an existing product. This isn't.
You'll be working on questions like:
How do you prove that licensed data produces better agent outputs than scraped data, with rigor that holds up to a customer's evaluation team?
What does a retrieval system look like when freshness, provenance, and rights are first-class signals alongside relevance?
How do you build evals that are honest enough to fail your own deploys and convincing enough to win customers?
This role sits at the core of whether Redpine is differentiated infrastructure or a nicer API on commodity vector search.
Practical details
This role is based in Stockholm and requires a valid Swedish work permit or similar eligibility
Relocation support available
Competitive salary and meaningful equity
About Redpine
If models were the first wave of AI, and compute the second, we're building the data layer that comes next.
Only a small fraction of the world's data is on the open internet. The rest, high-quality, domain-specific, often critical, sits behind paywalls, in databases, or with rights holders. Redpine is building the infrastructure to unlock it.
We provide AI builders and autonomous agents with access to licensed, high-quality, multimodal data through a unified platform and API. The goal is simple: make AI systems more accurate, more useful, and grounded in real-world information.
We're backed by Nordic Ninja, Node VC, and Luminar, alongside angels from OpenAI, Spotify, and Perplexity.
