Code barrier
- Requires Python/R
- Building from scratch takes 3–4 months
- Domain experts locked out
- No data advantage without years of engagements
binary classification · no code · GDPR ready
The score your model actually gets in production.
Most tools show you the optimistic number. We show you both — and our model learns from every engagement so the gap keeps closing.
v3.1.0 live · 174 tests passing · no vendor lock-in
The problem
Three barriers block every analytical question.
Code, compliance, and explainability slow down binary classification work before the modelling even starts.
Compliance barrier
- Cloud vendors need DPAs
- Sensitive data can’t leave org
- GDPR/HIPAA/Basel III friction
Explainability barrier
- Tools return a number, no audit trail
- EU AI Act requires transparency
- Regulators want why, not just what
What makes us different
A model that learns from every engagement.
The model trains on anonymised data from all engagements, human-reviewed, never autonomous.
model_core.yaml
read-only
model_families: [naive_bayes, logistic, random_forest, xgboost, lightgbm] learns_from_clients: true # anonymised, aggregated data_shared_between_clients: false # never self_updating: false # human review on every update evaluation: "nested_cv" output: "actionable_recommendations" audit_trail: true
Why faster
- Five families trained in one session
- No glue code — fully automated
- Model pre-informed by prior engagements
- Days from data handoff to output
Why safer
- Every run serialised to YAML
- Additive-only audit log
- Human-reviewed quality control
- Outputs labelled indicative, not decisions
How it works
Five steps. Raw data to actionable recommendations.
A controlled path from data readiness to reproducible output.
01
Consultancy
Data cleaning: on-site audit, anonymisation, leakage detection
billable02
Upload
Cleaned CSV loaded, screened
03
Evaluation
Target column, split strategy
04
Train & tune
Five families, nested CV
automated05
Deliver
Decision Letters, SHAP, audit trail
outputSecurity & compliance
Built for regulated industries — not retrofitted.
Controls are designed around data minimisation, reproducibility, human oversight, and honest scoring.
Data never shared
Your data trains our model for your project only, never exposed to other clients.
PII removed upfront
Identifiers removed, IDs hashed, sensitive attributes generalised before modelling.
Full audit trail
Dataset fingerprint, package versions, seed and config logged per run, reproducible offline.
Human quality control
We review every model update, no autonomous retraining.
Regulation-ready
GDPR, HIPAA, Basel III, EU AI Act addressed by design.
Honest scoring
Nested CV corrects for selection bias, both scores shown side-by-side.
Capability comparison
Where competitors stop — and we continue.
A focused view of service capabilities for regulated binary classification work.
| Capability | DataRobot | Azure AutoML | H2O | Data Owl |
|---|---|---|---|---|
| No-code training | ✓ | ✓ | ✕ | ✓ |
| On-site data cleaning | ✕ | ✕ | ✕ | ✓ |
| Nested-CV honest score | ✕ | ✕ | ✕ | ✓ |
| Model improves per engagement | ✕ | ✕ | ✕ | ✓ |
| Human quality control | ✕ | ✕ | ✕ | ✓ |
| Per-row Decision Letters | Enterprise only | ✕ | ✕ | ✓ |
| Full reproducibility (YAML + log) | ✕ | ✕ | ✕ | ✓ |
Based on public documentation, 2026.
FAQ
Seven answers before the first call.
Direct answers on data sharing, protection, regulation, model control, build time, and deliverables.
Is our data shared with other clients?
No. Your data is used exclusively within your engagement. It contributes to improving the model in fully anonymised form only — never exposed to, accessible by, or traceable to any other client.
How do you make sure our data is protected?
Before any modelling takes place, direct identifiers are removed, customer IDs are hashed, and sensitive attributes are generalised. The model only ever processes anonymised data. We retain the minimum necessary for the analytical question.
How do you handle GDPR?
Data minimisation at source: identifiers removed, IDs hashed, sensitive attributes generalised before modelling. In line with GDPR Article 5(1)(c). We do not transfer data to third parties.
What other EU regulations do you address?
The EU AI Act requires transparency and human oversight for AI-assisted decisions. Our outputs include per-row SHAP explanations and Decision Letters that document the reasoning behind every recommendation — suitable for regulatory review. We also address HIPAA and Basel III model risk requirements.
Who controls the model? Can it learn on its own?
We do. The model does not self-update or retrain autonomously. Every update is reviewed and approved by our team before deployment — a deliberate design choice to guarantee consistent, predictable quality.
How long would it take to build this ourselves?
A programmer building a comparable pipeline from scratch would need 3–4 months: setting up model families, writing evaluation logic, running tests, iterating on bugs. And they still wouldn’t have the accumulated data advantage that comes from years of real client engagements.
What do we receive at the end?
Actionable recommendations — not a model to maintain. Per-row Decision Letters, SHAP-based driver explanations, and a full audit trail (YAML config + run log) ready for internal review or regulatory submission.
Services
From raw data to actionable recommendations.
Data cleaning consultation
Data upload & screening
Evaluation plan
Train & tune
Deliver recommendations
Contact