SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Zaimler is building a platform to democratize enterprise data discovery and preparation for the generative AI era. The company bridges the gap between business knowledge and data by making data AI-ready faster and more reliably, with a foundation of semantic knowledge that leads to more accurate models and better business outcomes.
In this role, you'll work on systems that transform messy enterprise data into semantic models that AI agents can reason over reliably. The core challenge is not just calling an LLM—it's defining what "correct" means in unfamiliar customer domains and proving you've achieved it. A knowledge graph that looks right but is subtly wrong is worse than no graph, because agents will act on it.
You'll own multiple surfaces of the ML stack: building and improving NLP and retrieval systems that extract structured knowledge from unstructured enterprise data; working on the LLM layer (prompting, fine-tuning, RAG architectures); improving the retrieval stack (semantic search, vector storage, hybrid approaches, reranking); and building evaluation frameworks that define domain-specific correctness and catch regressions before customers do. You'll also maintain production pipelines that ingest, process, and serve this data at scale, and collaborate closely with customers, product, and platform engineers to ensure your work solves real business problems.
Success looks like shipping extraction or retrieval improvements in your first weeks, owning a real ML stack surface within months, and becoming the person the team routes new customer domains to because you consistently deliver working solutions.
The team is small (four ML engineers and a director) with deep expertise from LinkedIn, Visa, Truera, Hive, and Branch. You'll work on genuinely unsolved problems with real enterprise data across insurance, travel, and technology—not benchmarks. The company is backed by top VC funds and is signing enterprises faster than they can model their data.
