Summary
✨ AI‑Generated
An AI Developer role focused on owning LLM-powered features end to end, from prompts and prototypes to production outcomes. You will build product-facing features across the stack, evaluate and optimize multiple language models for quality, cost, and latency, implement traceability, and design human-in-the-loop agent workflows.
Highlights
End-to-end ownership of LLM-powered features, from prototyping through production. The role offers hands-on work with multiple leading language models, measurable evaluation loops, detailed tracing and prompt provenance, and human-in-the-loop agent workflows.
Description
Your mission
You own LLM-powered features end to end, from the prompt to what the editor sees, and you care whether the output actually gets published.
You get ideas into working shape quickly instead of waiting for the perfect spec, and you stay with them until they hold up in production.
Build product-facing features across the stack, with a clear bias toward shipping and iterating.Work on our multi-LLM pipeline (Anthropic, OpenAI, Gemini, Fireworks/Kimi) and make deliberate calls on model selection, cost and latency.Build eval loops against real editorial output: define the baseline, iterate, and be able to show what measurably improved.Own traceability across our LLM systems, meaning per-step tracing and prompt provenance, so a bad output can be traced back to the call that produced it.Design human-in-the-loop agent workflows: agents propose, the newsroom decides.
Nothing goes live autonomously.Prototype fast, harden afterwards, and ship your own work to production, including configuration, feature flags and scheduled jobs.Bring your own product ideas, rather than only working through tickets.
Your profile
Must-have:
Demonstrated product instinct and speed: you ship working things and make pragmatic trade-offs.AI-native daily development (Claude, Cursor) as a real workflow, not a demo.Hands-on experience with LLM orchestration (function calling, agents, evaluation).You have built an eval loop, not just written prompts: baseline, iteration, measurable result.You have traced LLM behaviour in production (LangSmith or equivalent) and used it to find a real bug.You can explain when a deterministic pipeline is the better choice than an agent.Python, and the willingness to own the deployment of your own work.Fluent English
Nice-to-have:
Depth in FastAPI.
We will also look at it if you have shown something comparable in another language.Enough frontend confidence to own a feature end to end.Experience with publishing or editorial workflows.German.
Why us?
Direct collaboration with the founders and real influence on product and cultureAn AI-native working environment with first-class toolsFlexible hybrid setup with an office in central Berlin and remote daysCompetitive salary and equity optionFlat hierarchy, short paths and fast decisions
About Us
BeatSquares transforms large content archives into new formats to reach audiences.
A single story can now reach everyone.
Our mission is to bring trusted journalistic information and factual content exactly where people need it—through a scalable, modern platform.
