# Limebit

> Limebit is a reviewed AI agency in Berlin, Germany, listed in the AIAgencies.eu register and last assessed 2026-09-09.

Canonical page: https://www.aiagencies.eu/agency/limebit
Generated: 2026-09-28 · Curated by Gabor Kiss · No paid placements — inclusion, score, and rank are never sold.

## Facts
- Location: Berlin, Germany
- Website: https://www.limebit.de
- Hourly rate: not published
- Team size: 2-9
- Founded: 2016
- Services: AI Development, AI Consulting
- Industries: Healthcare
- LinkedIn: https://www.linkedin.com/company/limebit

## About
Limebit is a Berlin data science and machine learning firm working on real-world evidence, health-insurance routine data and causal methods for pharma and healthcare clients.

## Documented clients
Documented engagements: AstraZeneca, Amgen, Bayer, and Roche. 2 of 4 carry a source URL on file.

## Curator assessment
Reviewed 2026-09-09 by Gabor Kiss against published evidence.

A health data science firm rather than a general AI shop, and the reference page reads like a research group's: client, data source, objective, method and focus on every entry.

The strongest is the synthetic-data work for AstraZeneca and Amgen across 2023 and 2024, training deep networks on a German health insurer's data to produce a synthetic set with the same statistical properties and then evaluating fidelity against privacy. Alongside it, machine-learning validation of electronic prescriptions has run in a Kubernetes environment for German health insurers from 2022 to 2025, and a breast-cancer registry analysis for an international pharmaceutical company looks for early indicators of metastasis.

Three commitments are published and they are the reason to take the firm seriously: every figure reproducible from a versioned pipeline rather than a notebook, consultants who read study protocols and statistical analysis plans as a matter of course, and full transfer of code, models and documentation at the end with no license attached.

The weaker fit is a buyer outside health and life sciences, or one who needs an outcome figure before signing, because the references publish method rather than result.

### Key strengths
- Full transfer of code, models, documentation and training artifacts at project end, stated as a principle rather than negotiated—no vendor lock-in
- Every deliverable reproducible from a versioned pipeline, published as a commitment against one-off notebooks
- Regulated-domain fluency is specific: real-world evidence, causal inference, Bayesian workflows and cohort phenotyping on statutory health-insurance and registry data

### Good to know
- The references publish method and objective but no result—ask what the retaxation and prescription models detect and at what rate
- Bayer, Roche, DAK and the other names carry no described engagement; the written-up work is AstraZeneca, Amgen and anonymized insurers
- A team of two to nine in a domain where projects run for years—ask who covers the engagement if a key person leaves

---
Source: AIAgencies.eu — the curated register of European AI agencies. Rate provenance: "agency-confirmed" = disclosed via submission or claimed profile; "directory-listed" = Clutch or two agreeing B2B directories (lone unverified sources are never written). Full method: https://www.aiagencies.eu/methodology · Rate index: https://www.aiagencies.eu/rates