Can expertise scale safely?
Turn fragmented track evidence into useful coaching while preserving the coach's judgement, voice and family relationship.
Portfolio · AI-enabled service venture
Unleash.
Intelligence in your corner.
Unleash is a platform for drivers and the people supporting them. Its karting offering, Unleash Racing, connects session evidence and learning. Existing coaching workspaces produce coach-reviewed debriefs; the family and adult-driver experience is forthcoming.
Private product evidence · repository main at a5b103d · reviewed 4 October 2026 · brand updated 11 October 2026 · public website and access checked 12 October 2026
Executive view
A service experiment in preserving expert authority.
Unleash is deliberately outside MiddleLeap's regulated core. Its relevance is the operating principle: AI may increase the reach, memory and consistency of an expert service without becoming the accountable expert.
In pilot use
The product assembles evidence, drafts a grounded debrief in the coach's voice and carries each driver's history into the next session.
The coach decides
AI cannot issue advice; a coach edit and sign-off are required, and a sent debrief cannot be changed afterwards.
Human-in-the-loop design
The experiment informs service propositions where AI assists perception, synthesis and memory but cannot own the decision.
Guided trials
Open guided, one-driver trials with agreed scope, then prove repeat usage across more coaches and teams before wider launch.
01
Why it exists
Elite coaching insight should not depend on an elite coaching budget.
Karting produces rich evidence—telemetry, video, photographs, voice notes, setups and weather—but turning it into useful, personal feedback is slow, and most of it is forgotten by the next race weekend. Unleash makes that synthesis repeatable and keeps the memory, without replacing the coach.
Evidence is scattered
Track-day observations arrive in different formats and often disappear before they become actionable learning.
One coached debrief, same day
Capture once, assemble the day, draft a grounded debrief and return it to the coach for judgement before it reaches the family.
The coach is the product
AI is invisible to the family. It supports perception, synthesis and recall; it cannot send advice or claim facts outside the evidence.
02
Product architecture
A traceable path from track evidence to coach-owned advice.
The runtime is a staged, durable pipeline. A consent gate runs first; vision, transcription and synthesis then create structured findings, enriched with the driver's own history, before final prose is written. A deterministic grounding and confidentiality check sits before the coach review and delivery boundary.
PeopleCoach + familyCapture · context · judgement
EvidenceMedia + telemetryPhoto · video · voice · setups · weather
ExperienceNext.js PWAOffline trackside capture and review
Stage 1Vision + WhisperServer-side extraction · class-aware terms
Stages 2–3Findings + proseDriver memory · coach voice · grounding
Human gateCoach edit + sign-offMandatory · sent debriefs locked
System of recordSupabase + RLSPer-team isolation · consent · memory
OutputPDF · email · WhatsAppFamily portal by link, no accounts
03
Loom-informed delivery
Start with the coach's gold standard, then make every AI step answer to it.
Unleash applies the Loom's evidence, specification and human-authority principles to venture building. This is disciplined use of the method's ideas—not a claim of full regulated-harness adoption.
Observe
Coach workflow, track-day evidence and a gold-standard debrief
Frame
Human-led product principles, consent boundaries and acceptance criteria
Design
A design-system component library and a capture-to-debrief architecture
Build
Claude Code loops with server-side AI and repository tripwires
Evaluate
Gold-set and live evals, RLS checks and mandatory coach sign-off
Technical build recordAI systems · Technology · Quality controls+
04
AI systems
AI drafts and remembers. The coach decides.
Claude Code builds the product; Anthropic Claude and OpenAI Whisper run inside it, server-side only. Each model tier is one switch in a registry, so a model change is verified by the evaluation lane before it reaches a real debrief.
Staged, server-side
Sonnet reads telemetry and photos, Whisper transcribes voice notes primed with corner and class-specific engine terms, Sonnet structures findings and Opus writes the prose.
Learned from edits
A voice profile is distilled from each coach's own edits. A coach can direct a rewrite of one section, returned as a proposal and never written automatically.
Structured, not guessed
Open and resolved focus threads, milestones, homework follow-through and lap trends feed each debrief, with a recall card when a driver returns to a track.
Deterministic check
Every number and corner in the prose must exist in the structured findings. The check is code, not a model marking its own work.
Right content, right reader
Exact setup figures stay with the team, focus areas are capped at three, and tone adapts to the driver's age on the session date.
Measured, not captured
Haiku matches voice notes to drivers. Cost, latency and errors are logged for every call; prompt and completion text are never captured.
Consent gate→Server-only AI→Grounding check→Coach edit→Human send
05
Technology
A trackside PWA backed by a secure, observable AI pipeline.
TypeScriptNext.js 15 + React 19Offline PWASupabase PostgresSupabase Auth + RLSSupabase StorageAnthropic ClaudeOpenAI WhisperHeadless Chromium PDFResendWhatsApp Cloud APIPostHogSentryVitest + PlaywrightpgTAPVercel
06
Quality system
Quality is measured at the advice boundary.
Build integrity
Types, lint, unit and component tests, database policies and the offline evaluation harness run on every change.
AI containment
Repository checks prevent browser-side model calls and committed secrets; a separate secret scan runs in CI.
RLS integrity
Migrations are replayed and pgTAP suites test that each team and coach can only reach its own data.
Grounded output
A gold set scores every change; a weekly live eval checks grounding and drift. The gold set is still small and mostly synthetic.
Synthetic track day
Each week a whole track day is driven through the real pipeline in an isolated sandbox organisation and verified end to end.
Coach authority
A coach edit and sign-off are required before a debrief becomes advice, and the database keeps sent debriefs immutable.
07
Development record
The core loop and memory are live; expansion remains evidence-led.
One-loop proof
Proven
Core debrief
Live
Operations spine
Live
Memory + coach voice
Live
Coach workbench + delivery
Built, gated
Wider market
Roadmap
Known boundary
Existing coaching accounts can sign in at unleash.training/login. The public Unleash Racing website explains the proposition and collects early-access interest; the family and adult-driver experience is forthcoming, and availability and onboarding are confirmed separately. The technical record on this page describes the coaching product reviewed on 4 October 2026. Production confidence still depends on repeat usage across more coaches and teams, live messaging credentials and a larger real-world gold set. Features built behind flags are not claimed as live. Parent consent and coach authority remain human responsibilities, not model decisions.
08
Evidence register
Claims are grounded in a reviewed private build record.
The repository is not publicly accessible. This register names the material reviewed; detailed source access can be considered during appropriate diligence.
R1Repository overview and product principlesReviewed
R2Current-state architectureReviewed
R3Product requirementsReviewed
R4Technical architectureReviewed
R5RoadmapReviewed
R6Model registry and AI pipelineReviewed
R7Continuous-integration workflowsReviewed
R8Deployment runbookReviewed
What this venture proves
AI can deepen a human service without taking authority away from the expert.
Bring the product, operating-model and governed-AI learning into your service proposition.