Skip to content

MVP scope (Phase 1 — "Seed")

Status: proposed. The MVP is the smallest thing that proves the core promise: point a camera at a plant and learn something true, in your language, safely.

In scope

  1. Plant identification — photo → ranked candidates with honest confidence and look-alike warnings (server-assisted first).
  2. Species pages for a curated starter set (~300 culturally and medicinally relevant species), seeded from open taxonomy with every field sourced.
  3. Names in ≥ 3 South African languages with attribution (candidates: isiZulu, isiXhosa, Tshivenda — to be confirmed with custodians).
  4. Traditional use claims — only sourced and safety-reviewed; shown as "traditionally used by…".
  5. Contribute — add a name, a use, a recipe or a note, with a plain-language consent step (visibility, attribution, AI-use).
  6. Offline use for downloaded packs; sync when online.
  7. Safety model v1 — toxicity/look-alike warnings, location blurring, disclaimers.
  8. Basic review queue for curators and safety reviewers (web).
  9. Localised UI in the MVP languages and English.

Explicitly out of scope for MVP

  • Open-ended AI chat (launch with retrieval-only "ask about this plant" or none)
  • Audio capture and ASR (Phase 2)
  • iOS and web consumer app
  • Social features, comments, feeds
  • Sacred/custodian tiers beyond basic support (design for them; limit launch content to public and community)
  • Monetisation of any kind

Success criteria

  • Top-5 identification accuracy on the SA evaluation set at or above an agreed threshold, with calibrated confidence.
  • 100 % of published claims have a source and a consent record.
  • Zero un-reviewed toxic-species use claims published publicly.
  • Core flows usable offline on a 2 GB Android Go device within the performance budgets.
  • At least 3 language custodians have reviewed their language's UI and glossary.
  • Qualitative: knowledge holders testing the contribute flow say it feels respectful and in their control.

Key risks

Risk Mitigation
ID accuracy too low on SA flora Spike first; narrow to a curated species set; abstain more
Not enough quality community content at launch Seed from published/open sources + partner programmes; invite holders early
Consent UX too complex Co-design with holders; voice-guided consent
Funding Early grants/partnerships; keep infra cheap
Legal uncertainty Counsel review before accepting community content