AI built around facts,
not black boxes.
Raw telemetry stays a deterministic engineering problem. AI is used where it adds value: explanation, prioritisation, cross-context reasoning and communication.
Three capture surfaces. One web app. One source of truth.
Bosch .bmsbin + VBOX telemetry → verified facts
Driver voice notes → structured session feedback
Uploads, typed notes, setup sheets and review
API + UI · deterministic insight engine · structured AI calls
Design intent: Australia East, in-region tenant, with raw telemetry retained locally and a compact processed fact set uploaded.
Different models for different jobs
The line the AI will not cross
The AI amplifies the engineer's judgement. It never replaces it.
Human-in-the-loop
Nothing reaches the driver or car without engineer approval.
Advisory only
The system offers best-supported options and why. It never actions the car.
Every claim cited
Evidence links back to laps, corners, channels and fact IDs.
Honest uncertainty
Low confidence is stated plainly and conflicting feedback is surfaced.
Self-calibrating
Predicted versus actual can be tracked to report the system's own hit rate.
In-region, authenticated and auditable
Microsoft Entra ID
SSO, role-based access and managed identities for application and service access.
Azure Key Vault
Centralised key management with encryption at rest and in transit.
Australia East residency
The design specifies in-region data residency with telemetry/IP retained within the team's environment.
Azure Monitor
Health, pipeline latency, model usage and operating cost visibility.
Audit trail
Recommendations, evidence and engineer decisions are logged for accountability.
Grounded retrieval
Season history, setup references, car manual and regulations can be retrieved before suggestions are shown.