prddraft
PRD: Curator pilot MVP — trustworthy session loop
Overview
Problem Statement
Coaches lose hours to manual, inconsistent after-session work and lose continuity between sessions; generic AI output isn't reproducible or client-safe. (See 01-problem-framing.md.)
Solution Summary
A coaching-native pipeline that turns a consented session into a reproducible, evidence-linked analysis and an editable client follow-up, with depersonalisation and human-in-the-loop, plus a cumulative client dossier.
Target Users
Practicing ICF-oriented coaches with 6–15 active clients (primary user + likely buyer). Secondary beneficiaries: their clients, program leads/sponsors.
Goals & Success Metrics
Goals
- Prove coaches trust and send the AI-drafted follow-up.
- Prove the analysis is reproducible and human-comparable.
- Confirm the after-session pain is real and time is saved.
Success Metrics
| Metric |
Baseline |
Target |
Timeline |
| Trusted follow-ups sent / active coach / week (NSM) |
unmeasured |
establish + grow |
pilot |
| Analysis reproducibility (F1 vs gold; inter-run agreement) |
unmeasured |
F1 ≥ 0.70 |
pilot |
| Coach minutes/session (before vs with) |
unmeasured |
meaningful drop |
pilot |
| Coach-rated usefulness |
n/a |
≥ 8/10 |
pilot |
Non-Goals
- Live in-session suggestions.
- Mobile client app; coach marketplace.
- Billing automation; non-US compliance.
User Stories
| ID |
User Story |
Priority |
| US-1 |
As a coach, I want a speaker-separated transcript of a consented session so I have an accurate base |
P0 |
| US-2 |
As a coach, I want a structured, evidence-quoted analysis so I trust what I read |
P0 |
| US-3 |
As a coach, I want an editable draft client summary so I send useful follow-up fast |
P0 |
| US-4 |
As a coach, I want sensitive data depersonalised before AI so I protect my client |
P0 |
| US-5 |
As a coach, I want a cumulative client dossier so I keep continuity between sessions |
P1 |
| US-6 |
As a coach, I want an alert when a session goes out of coaching scope so I decide on referral |
P1 |
See 14-deliver-user-stories.md for acceptance criteria. |
|
|
Scope
In Scope
- Consent capture, ASR + diarization, depersonalisation, reproducible analysis with quotes, editable draft summary, minimal dossier.
Out of Scope
- Live suggestions, mobile app, marketplace, billing automation, clinical features.
Future Considerations
- Deeper dossier + micro-tracking — defer until trust/adoption proven.
- Pricing/packaging + payer decision — open question (see
18-pricing-packaging.md).
Solution Design
Functional Requirements
Intake & safety
- FR-1: Capture per-session consent before processing.
- FR-2: Depersonalise identifiers before any external LLM call; re-identify only in protected perimeter.
Analysis
- FR-3: Produce speaker-separated transcript.
- FR-4: Produce topic/insights/commitments/next-steps, each with ≥1 transcript quote.
- FR-5: Analysis is reproducible within a defined noise band (measured).
Delivery & continuity
- FR-6: Editable draft client summary; explicit coach "send" (no auto-send).
- FR-7: Per-client dossier accumulates insights/commitments + status.
User Experience
Web app: dossier-first pre-session view → upload/connect recording → review analysis (quotes inline) → edit draft → send. Human-in-the-loop is visible at every gate.
Edge Cases (summary; full list 16-deliver-edge-cases.md)
| Scenario |
Expected Behavior |
| No consent |
Run without transcription; no analysis |
| Out-of-scope/clinical signal |
Alert coach; system does not act |
| Poor audio / overlap |
Flag low-confidence diarization |
Technical Considerations
Constraints
- Reproducibility + depersonalisation are architecture requirements (ADR-001).
- US-first compliance (CCPA/CPRA).
Integration Points
- ASR/diarization vendor; LLM provider(s) with switchability; calendar/video; messenger for follow-up.
Data Requirements
- Consent records; retention policy (audio/video default 90 days, configurable per studio page); access logging.
Dependencies & Risks
Dependencies
| Dependency |
Owner |
Status |
Impact if Delayed |
| ASR vendor |
Eng |
TBD |
Blocks all |
| Pilot coaches |
Founder |
TBD |
No validation |
| Legal review |
Founder |
TBD |
Compliance |
Risks
| Risk |
Likelihood |
Impact |
Mitigation |
| Reproducibility unattainable |
M |
H |
Experiment-gate; pivot if fails |
| "Good enough" generic AI kills WTP |
M |
H |
Prove trust + continuity moat |
| Pain is nice-to-have |
M |
H |
Diary study baselines |
| Payer ambiguity |
H |
M |
Pilot pricing discovery |
Timeline & Milestones
| Milestone |
Description |
Target Date |
| Pilot build |
Trustworthy session loop |
TBD |
| Reproducibility experiment |
Hit F1/agreement target |
TBD |
| Pilot run (3–5 coaches) |
Adoption + trust signal |
TBD |
| Go/No-go |
v1 decision vs pivot |
TBD |
Open Questions
- [ ] Who pays — coach, client, or sponsor? Owner: Founder
- [ ] Is F1 ≥ 0.70 the right client-safe threshold? Owner: Founder/Eng
- [ ] Which ASR vendor for the target language? Owner: Eng
Appendix
01-problem-framing.md, 09-product-strategy.md, 10-metric-design-experimentation.md, 11-develop-adr.md, 12-specification-writing.md
Revision History
| Version |
Date |
Author |
Changes |
| 1.0 |
2026-06-18 |
PM pipeline |
Initial draft |