Avalon Build Package
MVP Defined
Where to start, what MVP 0.0 means, and how we decided what is in versus out. MVP 0.0 is Phases 0-1: 26 features, 3 months, $40-60K. It proves the engine before the full app is built around it.
The MVP Thesis
Prove that AI can produce believable, personalized hair style visualizations with enough fidelity that a real user would act on them—then wrap that engine in just enough app for an early user to complete the full experience: take a photo, browse styles, see their visualization, and hand off to a stylist.
01
Where to Start
The MVP is Phases 0 through 2. But where you start inside the MVP is a different question, and that is the right question to ask.
Everything in Phases 0 and 1 exists to answer one question before Phase 2 commits real time and money building an app around it: does the AI visualization engine work well enough to build a product around? That is the risk to de-risk first. The app is straightforward once the engine is proven. The engine is not guaranteed.
0
MVP 0.0 — Weeks 1-4
Consent Foundation
Legal infrastructure and the prospect portal. Illinois BIPA and CCPA compliance must exist before a single face image is processed. Non-negotiable. Four features.
MVP 0.0 start
1
MVP 0.0 — Months 1-3 — $40-60K
Engine + Infrastructure
Hair segmentation, Face Lock, fidelity gating, the ArcRouter integration, the image corpus pipeline. This phase answers the question. If it answers well, Phase 2 proceeds with confidence.
MVP 0.0 gate
2
Full MVP — longest phase
Consumer App + Beta
The PWA, capture flows, style catalog, feasibility engine, stylist directory, safety layer. By end of Phase 2 the product is usable end-to-end and ready for a closed beta.
Full MVP complete
The practical answer to "where do I start"
MVP 0.0 = Phase 0 + Phase 1. Target: 3 months, $40-60K in engineering. Phase 0 first: consent infrastructure and this portal. Four features, two to four weeks. Without it you cannot legally process a face image. Then Phase 1: the engine. This is where the real work and the real risk live. If the engine hits quality bar, Phase 2 (the full consumer app, the rest of the MVP) follows with confidence. If it does not, you learn that before committing the larger Phase 2 budget to an app built around a broken core.
02
How We Decided What Is MVP
Every feature in the matrix was put through three questions. A feature is MVP if it passes at least one. A feature is post-MVP if it passes none.
1
Does the core flow break without it for a beta user?
A beta user needs to: take a photo, select a style, receive a visualization, read a feasibility assessment, and optionally hand off to a stylist. If removing the feature prevents any of those steps, it is MVP.
✓ MVP
Skip to next question
2
Is there a legal or safety reason it must exist before launch?
Consent architecture, age gating, data subject rights, safety filtering, skin-tone bias monitoring. These are not features—they are operating conditions. The product cannot ethically or legally exist without them.
✓ MVP
Skip to next question
3
Does it de-risk the core technical question?
The engine (segmentation, Face Lock, fidelity gating, quality measurement) exists to answer whether the AI visualization works before the app is built around it. Everything that helps answer that question sooner is MVP.
✓ MVP
Post-MVP
03
What MVP Proves and What It Does Not
The MVP is not the finished product. It is the version that tests the idea with real users before full investment. It is designed to generate evidence, not revenue.
What MVP proves
- The AI visualization engine produces results users find believable and useful
- The feasibility layer adds real information a stylist would actually use
- Users complete the core flow (photo to visualization) without confusion
- Early stylists find the consultation handoff valuable enough to engage with
- The legal and safety foundation is sound before broader launch
What MVP does not prove
- Whether the stylist marketplace scales or converts to bookings
- Whether a commerce layer or subscription model is the right monetization
- Whether the engine is fast and cheap enough at scale (Phase 3+ concern)
- Whether native iOS and Android apps are better than the PWA
- Edge case quality: 360 previews, fine-tuned segmentation, baby hair treatment
04
What a Successful MVP Looks Like
Before committing to Phase 2 you want evidence from Phase 1 that the engine clears the bar. Here is what that looks like in practice.
✓
Fidelity gate passes on a diverse held-out eval set
The engine produces visualizations that pass the automated fidelity check across a representative range of hair types, skin tones, and style categories. The eval set (feature P-11) is built in Phase 1 specifically so this measurement exists.
✓
Stylist review of sample output says "I would use this"
Before Phase 2 starts building the full app, a small review session with working stylists validates that the output quality is high enough to stake a professional recommendation on. This is human validation of what the fidelity gate measures automatically.
✓
Capture rejection rate is within acceptable range by skin tone
Feature P-12 monitors whether the photo quality gate rejects photos at significantly different rates across skin tones. If it does, that is a signal to fix segmentation before Phase 2, not after users experience it.
05
The Full Detail
This document answers the starting question. The documents below carry the full scope, timeline, and budget detail.