PRODUCT PROPOSAL · MENTOR PREMIUM · 2026-07-21

Generate Portfolio.

Byron's own front-door apparatus — the card stream, the generated art, the voice, the proof locker — turned into a one-tap export every Mentor user gets. One URL that is them.

Exhibit A — a mentee's portfolio, built for this proposal

◆ Live demo · fictitious persona

Maya Reyes-Okonkwo

A jazz pianist re-skilling into audio engineering. Her bio, her sound (a minute of piano), her first mix, her journey milestones, her 12-week Mentor plan — and a plain résumé for the busy human. Voice narration in a synthetic voice. Built from a card dataset, art pipeline, and TTS pipeline — the same machinery as sherpadude.

Open the living portfolio → Or the ⚡ dense version: maya-demo.arnao.ai/resume · the per-user data model (data.json)
  • 8 cards — bio, 2 audio proofs, a live-set placeholder, 2 journey milestones, the Mentor plan, a résumé card
  • Generated art — one stylized illustration per card (no real person)
  • Synthetic voice — narration per card, voice "sage" (not a clone)
  • Proof locker — playable audio (piano + a full mix)
  • Work mode — collapses to a dense one-line list / résumé
  • Idle screensaver — auto-tours the cards with captions
  • Disclosures — demo-persona + AI-voice + AI-art, in the footer

This demo is a single-tenant, hand-built instance — I generated Maya's assets directly. The honest delta between it and a real product (multi-tenancy, onboarding, quotas, a moderation gate) is the build-out estimate below. That gap is the whole cost story.

The product · "one URL = who I am"

A PDF résumé is a dead artifact in an AI-saturated hiring market — skimmed by a model, forgotten. Mentor already holds the one thing a portfolio needs and no one else has: a mentee's goal, their 12-week plan, and their check-in history — a real record of growth. Generate Portfolio turns that private record into a public communication vehicle in a couple of clicks: the person, playable. Send one link and someone meets them — sound, proof, trajectory — with a dense résumé one tap away for whoever just wants the facts.

Why Mentor, not a generic site builder: the moat isn't hosting a page — Carrd and Behance already do that. It's the Mentor-data-to-portfolio pipeline: the mentee's goals and evidence are already structured inside Mentor, so the portfolio writes itself from work they've already done. The card-kit experience is the finish; the data is the wedge.

Free vs. premium — the split

✓ Free · every user

The résumé link

  • One clean URL — the dense, traditional one-pager
  • Auto-drafted from their Mentor goal + plan + profile
  • Print-friendly, loads instantly, no login to view
  • The "a non-AI human clicks it" mode — Byron's ⚡ work-mode idea
Marginal cost ≈ free (text + hosting). A retention hook: it makes the free tier useful to share.
★ Premium · the living portfolio

The full apparatus

  • Card stream with generated art + synthetic voice narration
  • Proof locker — upload audio/video/images as evidence
  • Journey milestones + live Mentor-plan progress card
  • Their own curation (reorder, edit, visibility, regenerate)
  • Screensaver / QR / share — a front door, not a page
  • Add-ons (metered): motion cards, "ask my portfolio" realtime voice
Marginal COGS ≈ $0.60–$0.90 one-time + pennies/mo. Trivially covered by any premium price (see costs).

How the apparatus maps — 1 owner → N mentees

Byron's apparatus todayBecomes (per mentee)Status
Global registry.jsonPer-user registry (their cards) — namespaced users/<id>/portfolio.jsonRESHAPE
Byron's /admin curationThe mentee curates their own cards — reorder, edit copy, toggle visibility & stageBUILD
Card-kit front-end
stream · x-ray detail · work-mode · screensaver
Same UI, themed per user — reused wholesale (this demo proves it)REUSE
Art pipeline
grok-imagine ~$0.05/img
Per-user card art on generate — same script, per-user prompts from their dataREUSE
Voice pipelineSynthetic TTS per card (not the clone — see safety). Same gpt-4o-mini-tts pathREUSE
stage / visibility tagsThe mentee's control surface — what's public, what stage each item is atREUSE
Vercel + GoDaddy deploy automationPath- or subdomain-per-user routing under one projectRESHAPE
—Onboarding: pull Mentor goal/plan/check-ins → draft cards → generate → user reviewsBUILD
—Quotas + spend guardrails; moderation gate; proof-locker upload/transcode; likeness/voice consentBUILD

Multi-tenancy — the routing call

arnao.ai DNS is one-level with no wildcard, and mentor.arnao.ai is owned by the mentor-proto project. So per-user subdomains (maya.mentor.arnao.ai) mean a wildcard CNAME + a router — real work for a vanity gain.

Recommendation: path-based v1 — portfolio.arnao.ai/<handle>, one project, zero per-user DNS, isolation by storage-key prefix + auth. Offer a vanity subdomain later as a premium upsell.

Storage & isolation

Per-user namespace in Vercel Blob (the store Mentor already uses): users/<id>/{portfolio.json, art/, voice/, uploads/}. Cross-tenant reads blocked by key prefix + the mentee's existing Mentor auth (Google GIS).

Critical line: a mentee's private Mentor data (goals, check-ins) stays private — only what they explicitly publish becomes the portfolio. Publishing is opt-in, per item.

Costs — "is it a couple clicks and free?"

Short answer: per-portfolio compute is under a dollar and hosting is pennies. The real cost is one-time engineering to productize, plus ongoing human moderation. Here's the honest breakdown.

Per-portfolio marginal COGS (one mentee, at generate time)

$0.05
per illustrated card (grok-imagine)
~$0.50
art · a ~10-card portfolio
~$0.03
synthetic narration, all cards (TTS)
~$0.10
LLM to draft bio/copy from Mentor data
~$0
hosting/bandwidth (static + CDN)
$0.60–0.90
total one-time COGS / portfolio

Optional metered add-ons (off by default): animated/motion cards ≈ $1–2 each (video gen); "ask my portfolio" realtime voice ≈ $0.20–0.60/session. Recurring: Blob storage ≈ a few cents/month per portfolio.

This demo's actuals: 8 generated illustrations $0.40 + synthetic narration (2,086 chars) $0.025 + music clips $0.00 (rendered locally) = ≈ $0.43 total, hand-built. At the productized ~10-card default it lands at ~$0.55–0.90.

Build-out — the honest engineering scope

WorkstreamWhat it isEst.
Tenancy + routingPer-user registry, storage namespacing, path routing, auth tie-in to Mentor GIS3–5 sessions
Generation flowMentor goal/plan/check-ins → draft cards → generate art + voice → user review/approve4–6 sessions
Mentee curation UITheir own admin: reorder, edit, visibility, stage, regenerate a card3–4 sessions
Quotas + spend guardPer-user generation caps, cost ledger, abuse throttle1–2 sessions
Moderation + consentScan generated + uploaded content; likeness/voice consent flow; takedown3–5 sessions
Proof-locker uploadsUser audio/video/image upload, storage, transcode, size limits2–3 sessions
Polish / QA / brandFront-door-grade polish, anti-AI-tell copy, cross-device2–3 sessions
18–28
build sessions (~3–5 focused weeks) for a real multi-tenant v1
~70%
of the front-end + pipelines reused from the existing apparatus
verdict
compute is nearly free; eng + moderation ops are the cost
Pricing coverage: at a plausible premium of $12–20/mo, one portfolio's sub-$1 generation + pennies/mo hosting is a rounding error. Even a heavy user regenerating monthly with a few motion cards stays under ~$3/mo COGS — gross margin stays high. The break-even math is dominated by the one-time build and the ongoing moderation headcount, not the model bills.

Risks & safety — before any of this ships

User-generated content on Byron's brand infra
A mentee could host something reputationally damaging on an arnao.ai property. Mitigation: a moderation gate on publish (automated scan + human-in-loop for flagged), a clear takedown path, and a neutral portfolio domain (e.g. a dedicated portfolio.*) so the brand isn't the host of record. Portfolios should not live on the same surface as Byron's own front door.
Likeness & voice ethics
Synthetic voices only by default — the demo narrates in a generic TTS voice and says so in the footer. A user's own cloned voice or a real photo requires a consent sample + explicit attestation; no cloning of anyone else, ever. Generated portraits are stylized, never photoreal of a real person.
Disclosure is non-negotiable
Even a synthetic voice saying "I'm Maya" is a mild deception surface. Every portfolio carries persistent AI-voice / AI-art / (where relevant) demo-or-real disclosures — matching Byron's standing rule on synthetic media. This demo models exactly that.
Moderation is real, recurring opex
Human review of flagged content is not $0 and not one-time. Scope it as ongoing cost from day one; do not price the feature as if compute were the only variable.
Scope / focus risk
Mentor is a coaching app; a portfolio builder is arguably a different product. Frame it as the natural output of coaching ("you did the work — here's the artifact"), not a pivot, or it dilutes the core. If it can't be that, it may not belong in Mentor.

Self-critique — Fable adversarial pass

The strongest case against building this

  • The pretty part is the cheap part. The demo makes generation look solved because it is — it's a $0.43 afternoon. The 80% you can't see (multi-tenant isolation, a moderation pipeline that survives a bad actor, consent flows with legal teeth) is where the weeks and the risk live. Don't let Exhibit A anchor the estimate low.
  • Is a portfolio Mentor's job at all? The wedge (Mentor data → portfolio) is real, but the market (Carrd, Read.cv, Behance, LinkedIn) is crowded and free-ish. If the differentiator collapses to "nicer cards," that's a weak moat.
  • UGC on arnao.ai is a standing liability. One offensive or fraudulent portfolio on a Byron-branded domain is a brand incident. The neutral-domain mitigation is necessary but adds cost and dilutes the "it's part of the brand" appeal.
  • Synthetic voice narrating a first-person bio is an ethics tightrope even with disclosure — it normalizes AI-spoken "me." For a founder whose brand is responsible AI, that's a scrutiny surface, not a freebie.
  • Moderation headcount is the real unit economics. The COGS table is honest about compute and quietly optimistic about people. At any real scale, human review dominates.

Where it survives the critique: as a premium retention feature (not a standalone product) whose free-tier résumé link is a genuine sharing hook, on a neutral domain, with moderation costed in from day one — it's a high-margin, on-brand extension of work mentees already did. The demo proves the experience is real and cheap to render. The open question is strategic focus, not feasibility.

Is this a couple clicks and basically free? — Yes on compute. No on people.
Per portfolio: <$1 to generate, pennies/mo to host. To productize: ~3–5 focused build weeks (70% reused) + an ongoing moderation gate. Recommend a path-based v1 on a neutral domain, free résumé link for all, living portfolio as premium. Decision points for you: (1) build or shelve, (2) neutral domain vs arnao.ai, (3) does "AI-spoken bio" clear your own brand bar?

Links