About Freshflow
Freshflow builds the AI operating system for the fresh food supply chain, starting with ordering and inventory optimisation for fresh departments. Grocery retailers across Germany and France, including EDEKA, REWE, Carrefour, Intermarché and E.Leclerc, use Freshflow every day to cut food waste and keep shelves full. Stores accept 93% of our order proposals, and we recently closed a $10M Series A.
Our forecasting and ordering engine already makes ordering decisions under uncertainty. The next step is making both our product and the way we build it AI-native. We want software that explains itself to a store manager, reads the messy files retailers send us, and handles operational work that today needs a human. We want that software to be written, tested and shipped by agents. We're looking for an AI Engineer to help build this.
The goal of this role
We believe the job of an engineer is moving from writing code to designing the system that writes it. In this role you won't spend your time writing production code by hand or reviewing agent-written code line by line. Instead, you build the agentic engineering workflows that take a feature from specification to production with high quality, end to end. You also build the oversight tools that let humans steer those workflows and trust what comes out. Success means agents deliver work we would gladly ship, and humans spend their time on direction, judgement and the workflows themselves.
What you'll do
- Build agentic engineering workflows end to end: agents that turn a specification into code, then write tests, run evals, review each other's work, fix what fails, and ship to production. You design the steps, the agents, their tools and the quality gates between them.
- Make quality something the system proves, not something a person checks by reading code: automated tests, evals, backtests, adversarial review agents and production monitoring that together decide whether a change is good enough to ship.
- Build the oversight tools for humans: dashboards, summaries, approval points and escalation paths. These let a person set direction, spot when a workflow is going wrong and step in at the right level, without reading the code.
- Deliver AI-driven product features through these workflows, into the hands of store staff and our customer teams. Examples of what's on the table:
- Explaining in plain language why we recommended an order, grounded in the forecast, inventory and cost data behind it.
- Turning unstructured retailer inputs (order guides, supplier files, item master data) into clean, structured data.
- Assistants and agents that help Customer Success and Operations investigate store issues and act on them.
- Keep improving the workflows: measure where agents fail, why, and what it costs, then fix the workflow rather than the output. Turn each lesson into a reusable skill, tool or check.
- Work across the company: partner closely with the rest of the Data Science team, Engineering, Product and Customer Success. Together, pick the problems where AI moves the needle for stores, and say no to the ones where it doesn't.
What you bring
This field is new, so years of experience matter less to us than what you've actually built. Senior or junior, we care about how well you can get agents to deliver.
- Deep hands-on experience with agentic software development: you already build software mainly by directing coding agents (e.g. Claude Code), and you have ideas about how to take humans out of the code-review loop.
- Experience building agents and multi-agent workflows: tool calling, planning, memory, orchestration, verification and guardrails.
- A track record of shipping LLM-powered product features that real users rely on, not just demos.
- A rigorous approach to evaluation and quality: building eval sets, LLM-as-judge, regression and integration testing, and measuring cost, latency and reliability. You know how to make a system prove its output is correct.
- Strong engineering judgement and fluency in Python. You know what good production software looks like, so you can define it for agents and recognise when they don't meet the bar.
- Solid SQL and comfort working with large, messy real-world data.
- Experience with cloud infrastructure, ideally GCP (BigQuery, Cloud Run, Vertex AI), plus Git, containers and CI/CD.
- A product mindset: you start from the user's problem and judge success by whether it's solved.
- Fluent English.
Nice to have
- Experience with MCP, designing tools and skills for agents, or building internal developer tooling.
- Experience with retrieval systems, structured extraction or document understanding.
- Exposure to forecasting, optimisation or supply chain problems.
- Grocery retail, perishables or B2B SaaS experience.
- German or French.
- Startup or scale-up experience.
What we offer
- Equity participation (VSOP).
- Work from anywhere in Germany, or hybrid from our Berlin office, with relocation support if you'd like to move to Berlin.
- A real-world problem with measurable impact: less food waste, fuller shelves.
- A small, senior team where you own your area from day one.
- A fast hiring process: screening call, case study and interviews, then an offer, typically within 2–3 weeks.
How to apply
Send your CV to jobs@freshflow.ai. If you have one, include a link to an agentic workflow or LLM-powered system you've built, and tell us how you made sure its output was good.
Freshflow is proud to be an equal opportunity employer. This means that we don't discriminate based on race or ethnic origin, color, the language(s) you speak, where you (or your parents) are from, whether or not you consider yourself to have a disability, your age, gender, gender identity, sex, sexual orientation, religion, beliefs, or political opinions. We're a diverse team in many ways, and we love it that way.