01 Context & Problem
The core user, a successful professional, is spiritually hungry but stuck: generic Western apps lack depth, Instagram healers can't be trusted, and the genuine teachers are unreachable from another country. The whole category fails on one question: is this person legit, and is this safe? The constraint isn't money, it's trust.
02 Role & Constraints
As AI Product Manager I owned the full product and engineering build: user, practitioner, and admin surfaces. Founder owns practice quality, verification judgment, and brand. Constraints: a lean team covering a marketplace's full surface area, a two-sided cold-start that must feel trustworthy on both sides, and a guiding principle from the spec, safety is product, not a compliance layer.
03 Product Approach
Verification is the spine. A practitioner can't list without passing a four-stage review (written application, video introduction, a full demo session scored on an eight-dimension rubric, and a twenty-session provisional window) for an aspirational sub-10% acceptance rate. Around it sits a full session lifecycle and a relationship model: after session one, a user has a Primary Practitioner, so the product compounds on continuity rather than one-off bookings.
Trust is the wedge: lead with verification the way fintechs lead with security. It's also portable: the trust layer is market-agnostic, so global expansion is go-to-market, not a rebuild.
04 Features Built
Four-stage verification
Written → video → scored demo → 20-session provisional. The moat.
Practitioner profiles
Verified badge, tier, lineage, intro video, ratings, availability.
Practitioner onboarding
Application plus a 7-step checklist gated before the first session.
Practitioner dashboard + CRM
Today / Schedule / Clients / Business, with per-client private notes.
User accounts + journey
Dashboard, reflective Journey surface, profile, and settings.
Search & filtering
By modality, language, tier, availability, price, and rating.
Booking flow
Three screens max: slot → account → payment, no surprises.
Stripe Connect payments
Multi-currency checkout and weekly automated practitioner payouts.
Embedded video (Daily.co)
In-app live sessions, no Zoom links, full control of the session.
Ratings & reviews
Opt-in, admin-moderated, feeding practitioner tier progression.
Trust & safety
Crisis-keyword detection with tiered response and country resources.
Admin / super-admin
Six sections: dashboard, users, practitioners, sessions, content, finance.
Also shipped: scoped messaging (intake + one post-session exchange), Google Calendar two-way sync, packages, gift sessions, referrals, and the content/practice library.
05 Architecture
Next.js (App Router) on Vercel; Supabase for Postgres + auth with row-level security separating user, practitioner, and admin roles. Stripe Connect runs marketplace payments and weekly multi-currency payouts; Daily.co powers embedded video; two-way Google Calendar sync drives practitioner availability.
06 Analytics & Observability
Instrumented from day one, so every funnel, drop-off, and error is visible, and so the data foundation the AI layer below depends on actually exists. Four tools, deliberately non-overlapping:
PostHog
Product analytics: funnels, event tracking, and session replay across the intro→paid and retention journeys.
Sentry
Error and performance monitoring: catches crashes and slow paths in checkout and live video.
Google Analytics
Acquisition analytics: traffic sources, landing-page conversion, and channel CAC.
Hotjar
Heatmaps and recordings: shows where users hesitate on profiles and at checkout.
07 AI Decisioning Layer
On top of the analytics stack above, the platform runs a tiered commission engine (30% → 25% → 20% as practitioners deliver more sessions) and market-specific pricing. That data foundation powers the AI decisioning layer.
Claude is integrated into the super-admin as a decisioning copilot reading the PostHog and analytics data: it recommends then automates commission and fee adjustments to protect margin, acts on analytics events (churn risk, timezone supply gaps, practitioner quality drift), and surfaces profit-optimization moves a human admin would miss. The sequencing was the point: the analytics substrate came first, and AI decides on top of it.
08 Status & Outcome
Full three-sided platform built across web and mobile; launch sequenced supply-first, with a Yoga Day campaign anchoring early demand.
50
Verified practitioners
1,200
Paid users
32%
Intro → paid conversion
70%
First → second session retention
09 Reflection / What's Next
Sequencing the verification, analytics, and admin foundation before any AI was deliberate: a system that adjusts real money is only as trustworthy as the data and review process beneath it. The platform is running successfully with the AI decisioning copilot live, and the same portable trust layer scales the model market by market.