The Team:
JobLeads helps millions of professionals across 40+ countries find their next role, matching them against millions of live jobs. We are building what we believe will be the best job search on the internet - one that understands what a candidate is actually looking for, even when they can't put it into a keyword. Search, matching, and recommendations sit at the core of our product, and applied AI is how we are already winning there.
Your Role:
As an Applied AI Engineer, you will build the intelligence behind how candidates and jobs find each other: semantic search, resume-based matching, LLM-driven query understanding, and recommendations.
The role evolves with the mission: whatever it takes to make our search the best job search on the internet is what you'll work on next. That makes out-of-the-box thinking a feature of the job, not a nice-to-have - we expect you to bring new approaches, methods, and frameworks to the table, whether they come from commercial model providers or straight from current research. If a paper published last month suggests a better way to retrieve, rank, or evaluate, we want you to be the person who spots it, tests it, and tells us whether it holds up on our data.
What stays constant is the way of working: form a hypothesis, benchmark it rigorously against the status quo, make it tangible in a prototype people can play with, ship the winner to production, and prove it in an A/B test. You'll work closely with the people who make the decisions and present your results to them directly - your benchmark can become a production experiment within days.
What You'll Be Doing:
- Design and improve retrieval systems over a large, messy real-world corpus at scale: embeddings, hybrid search, ranking, and re-ranking
- Shape the design of our vector search backbone: how jobs and resumes are represented, which embedding models to use, the right vector dimensionality for each purpose, etc.
- Build LLM pipelines for understanding queries, resumes, and jobs — extraction, enrichment, rewriting, matching - choosing the right model and technique for each step
- Own evaluation as a scientific discipline, not a checkbox:
- Curate benchmark datasets that capture both the failure modes you're fixing and the healthy cases you must not regress
- Design LLM-as-judge setups you can defend - ground and calibrate the judge against human sanity checks, detect its biases (a conservative judge will punish a correct query expansion for not being literal), and iterate on the judge itself when it measures the wrong thing
- Apply and adapt search-relevance methodology - precision@k, NDCG, pairwise preference testing - choosing metrics that resist gaming, and recognizing when a metric rewards degenerate behavior
- Read results like a scientist: cumulative vs. independent effects, why two valid metrics can disagree, what a distribution's tail says that its mean hides
- Quantify the full picture - quality, latency percentiles, and cost per query
- Build interactive prototypes so product managers and stakeholders can test your approaches hands-on before anything ships
- Partner with engineers to bring winners to production faithfully - no gap between what was benchmarked and what ships - and follow through into A/B test readouts
- Track the fast-moving model and research landscape and re-evaluate as new models, techniques, and papers appear, keeping us on the best quality-per-euro frontier
- See your work impact the way millions of job seekers find their next job in 42 countries
- Be surrounded by colleagues who understand you and help you grow in your role
What You'll Need:
- Degree in Computer Science, AI, or a related field from a top-tier program (e.g., TUM, University of Tübingen, Saarland University, RWTH Aachen, LMU Munich, TU Berlin, KIT, University of Freiburg, University of Bonn, FAU Erlangen-Nürnberg, TU Darmstadt, University of Stuttgart, Heidelberg University, HPI (Hasso Plattner Institute, Potsdam), TU Dresden, Uni Paderborn, or equivalent)
- Have hands-on experience building applied ML or LLM systems - retrieval, RAG, agentic pipelines, or traditional search - that real users touched
- Have genuine data-science depth: you design experiments, build baselines, and reason about metrics and their failure modes before trusting a number
- Are fluent in Python and comfortable owning a pipeline end to end: data, prompts, embeddings, retrieval, evaluation, and the glue in between
- Don't trust a demo: you build a baseline, a dataset, and a metric before declaring victory - and you notice when your metric is measuring the wrong thing
- Communicate clearly with non-ML colleagues and enjoy making your findings understandable, not just correct
- Are pragmatic and 80/20-minded: you'd rather ship a simple approach that captures most of the gain than a complex one that looks better on paper
Nice If You Have:
- A track record of taking methods from research papers into working systems - or contributions of your own (publications, open source, competition results)
- Experience with latency-sensitive production ML and knowing where in a product expensive calls are acceptable
- Experience running or analyzing A/B tests on search or recommendation systems
- Exposure to multiple model providers and cost/latency benchmarking across them
What You Can Expect On Board:
- One of the most exciting applied-AI roles in Germany. This is a rare chance to own the intelligence behind how millions of people find their next job — at real scale, with the latest tooling, and a short path from idea to production.
- You will never be bored - in the best sense. The problems are at the frontier of applied AI, the loop from idea to live experiment is short, and there's always a harder, more interesting question behind the one you just answered.
- The best tools, by default. You'll work with the latest AI tooling - Claude Code and whatever comes next - so you move fast, skip the busywork, and spend your energy where your potential actually shows.
- Friendly, kind, and fast - at a sustainable pace. We move quickly because the work is focused, not because anyone burns out. Communication is direct and warm, and decisions are made on evidence rather than hierarchy: benchmarks settle debates here.
- Amazing colleagues. Expect good conversations and good moments - experienced product managers, engineers, and data scientists who give real feedback, get genuinely excited about a clean result, and celebrate wins together.
- Memorable team experiences, from EU meetups to our annual company summer getaway.
We’d love to hear from graduates and PHDs of top-tier AI and Computer Science programs, particularly those with strong machine learning, LLM, or AI systems experience. This includes universities such as TUM, the University of Tübingen, Saarland University, RWTH Aachen, LMU Munich, TU Berlin, KIT, the University of Freiburg, the University of Bonn, and FAU Erlangen-Nürnberg - but outstanding candidates from other universities are equally encouraged to apply.