Dossier · AI / ML
Models with guardrails
Automation, extraction, assistants. Data and cost sit in the contract. We don’t sell “AI” on a contact form.
AI / ML on file
Cases where AI is a tool on an existing flow — not an invented product.
No Projects Found
Try adjusting your filters or check back later for new case studies.
AI protocol
From data to decision
First which data we’re allowed to use. Then which model. Then cost per call.
Data
Source, quality, legal basis.
Flow
Where the human is, where the model is, where fallback is.
Guardrails
Prompt, PII, cost cap, eval.
Measure
Errors, latency, cost. Not “wow”.
AI without process is a demo.
In production, the fallback when the model is wrong is what counts.
Portfolio questions
Q1.Do you train your own models?
When volume earns it. Otherwise: RAG / API with guardrails — cheaper and more controllable.
Q2.Is a site chat AI?
Only if it has data, fallback and measurement. Otherwise it’s a widget.
Same protocol
You have a repeatable flow. We’ll see if a model earns it.
Get a quote. Data and cost first, demo second.