← All work2026 · Data · Mobile

Ingredient Intelligence Engine

Messy label text in, canonical substances out — every resolution scored and sourced.

RoleSolo build — in progress
StackTypeScript · Deno · Supabase Edge Functions
Year2026

The hard part is the names

Scan a product label and you get chaos: INCI names, E-numbers like E300, Colour Index codes like CI 77491, French aliases like eau, and undisclosed mixtures like parfum. The engine resolves each token to a canonical substance and — the part most scanners skip — reports how it resolved it and with what confidence. Unresolvable tokens are marked as such, never guessed.

Health scoring follows a documented, source-cited methodology with an emphasis on endocrine-disruption evidence, personalized by user profile. The seed database is ~41 hand-curated substances, each individually sourced.

POST /resolve — responsejson
{
  "tokens": [
    { "raw": "AQUA",      "substance": "Water",           "method": "inci-alias",  "confidence": 1.0  },
    { "raw": "E300",      "substance": "Ascorbic acid",   "method": "e-number",    "confidence": 1.0  },
    { "raw": "CI 77491",  "substance": "Iron oxide red",  "method": "colour-index","confidence": 1.0  },
    { "raw": "parfum",    "substance": "Fragrance (undisclosed mixture)",
                           "method": "mixture-flag",       "confidence": 0.9  },
    { "raw": "xylotrol",  "substance": null,              "method": "unresolved",  "confidence": 0.0  }
  ],
  "note": "unresolvable tokens are marked, never guessed"
}
A resolution response — method and confidence per token

Honest scope

Phases 0–2 of 6 are complete and verified: the resolution engine works and is tested. Product ingestion, scoring UX, and the mobile app are the next phases. This page will grow with them.

Built with

Every technology this project runs on, and the job it does here. Each one links to the other projects it appears in.

Languages & runtimes

TypeScript↗Strict mode in every web and desktop codebase — typed data layers, discriminated-union reducers, generated database types, no implicit any.Deno↗Runtime for Supabase Edge Functions — the serverless API surface for several products.Python 3.11+↗Primary backend language: agent services, ETL pipelines, eval harnesses and CLI tooling. Typed end to end with Pydantic v2 models, checked under mypy --strict.

Backend & APIs

Supabase Edge Functions↗Serverless API surface; three separate internal FastAPI tools were migrated onto them and merged behind one console.

Retrieval, ML & vision

Entity resolution↗Mapping messy real-world tokens — INCI names, E-numbers, Colour Index codes, foreign-language aliases, undisclosed mixtures — to canonical substances, each with a resolution method and confidence; unresolvable tokens are marked, never guessed.OCR — Tesseract · ML Kit↗On-device text recognition for product labels, plus batch document tooling.

Mobile & desktop

Expo · React Native↗Scanning client with on-device text recognition and local-first data.

Databases & storage

Supabase↗Postgres, Auth (email OTP and OAuth), Realtime, Storage and Edge Functions as one backend behind both web and mobile clients.PostgreSQL↗19+ hand-written migrations, tenant provisioning and staff invites via RPCs, snapshot-sync triggers keeping derived views consistent, and generated types checked into the client.

Infrastructure & delivery

Docker↗Service images and reproducible local stacks brought up by a single command.pnpm workspaces · uv↗Supabase CLI and Deno pinned as dev dependencies — no global installs; one script starts Docker, applies migrations, seeds data, regenerates types, and runs functions and app side by side.

By the numbers

41Curated, cited substances
2 / 6Build phases complete