Hiring Now
Working Student AI / LLM · Berlin

A user asks a question. The database answers — reliably.
We build what's in between.

Think Spotify, but for enterprise data. Imagine a product manager at a streaming service wants to know:
"Do users under 25 listen to more podcasts or music on Monday mornings?"
The data exists — buried in hundreds of tables, millions of rows. Today, that question takes a data team and two days. With KEA, anyone just types the question and gets the answer. In seconds.

KEA is an AI-powered data intelligence platform. Our first clients are in market research — but the technology is domain-agnostic: any structured database, any industry. We turn natural language into verified database queries across fully isolated multi-tenant environments — each with its own data model, glossary, and search index. Thousands to hundreds of thousands of variables per tenant, on-prem ready, zero data exposure, role-based access, full audit trails. Most NL-to-SQL tools choke on real enterprise data. Ours doesn't — because we treat queries as pipelines, not prompts. Small team in Berlin. Weeks from launch. Paying customers lined up.

What You'd Own
  • Prompt engineering & LLM orchestration at the core of the product
  • Retrieval logic — making search precise, not just fast
  • Early weeks: improve query accuracy, harden edge cases, ship a feature end-to-end
  • Architecture decisions — you're in the room, not outside it
  • Push models to their limits, find where they break, fix it
How We Think
  • Turn chaos into structure — messy input in, clean pipeline out
  • Decompose everything into small, testable steps
  • Every decision traceable. Every output explainable.
  • Systems that self-correct and learn from failure
  • If it's not transparent, it doesn't ship.
Stack & Tooling
Production stack — individual components may evolve as we scale post-launch.
Python TypeScript LLM Orchestration Prompt Engineering DuckDB Supabase n8n Embeddings BM25 + Semantic Hybrid Search Cross-Encoder Reranking NL-to-SQL RAG Intent Routing Self-Correcting Generation Multi-Tenant Isolation Role-Based Access AI-Assisted Dev
You
  • Strong Python — the entire backend runs on it
  • Obsessed with LLMs, retrieval, or making AI actually reliable
  • Have built and shipped something — GitHub, prototypes, anything
  • Learn fast. Debug methodically. Take ownership.
  • SQL helps. Relentless curiosity is non-negotiable.
Us
  • Small team — no layers, no politics, no process theater
  • Biweekly architecture sessions, in person, with real decisions
  • We build with AI and review with humans — speed + rigor
  • Best argument wins. Titles don't.
  • Full-time down the road? Open to it — no pressure either way

You don't need to check every box — but we're looking for someone who's genuinely strong, not just interested. Exceptional in one or two areas, fast learner everywhere else. One real project beats ten online courses.

Berlin
Based here, not remote
12–20h
Per week, flexible
€25/h
Top of market
ASAP
Start
Full-time
Possible, not promised
Applications closed. Due to the overwhelming number of applications, we are no longer accepting new submissions for this role. Thank you to everyone who reached out — we'll be in touch with shortlisted candidates directly.
Closed
No new applications

KEA · Native Research · Berlin · 2026