ProductDataOps · Hong Kong · 2025 — Present
Rosetta data decoder
Founded a platform for structuring, enriching and governing complex catalogs across supply chain and commerce.
Context
Product data remains the most fragmented, manual and value-destroying layer in commerce and supply chain.
Intervention
- Owned product vision, semantic data architecture and workflow design.
- Ran positioning and early market validation with design partners.
Outcome
- Concept to working product and validated demand signal.
Rosetta — the decoder layer
Supply chains run on product data, yet most of it still arrives as spreadsheets, ERP exports, supplier PDFs and portal dumps that people clean by hand. Rosetta is a thin decoder layer: it reads whatever format arrives and turns it into the structures a business already trusts — matrices, specs, attribute types, taxonomies and supply-chain maps. It sits on top of existing systems rather than replacing them.
Raw AI can read a document. It cannot decide what a field means, which unit is acceptable, or when a value should be rejected. That judgement is what the layer adds — and every new dataset makes the next one easier to use.
Canonical naming
Every supplier names the same spec differently. Each variant maps to one canonical field, so attributes stay comparable across suppliers and documents.
Unit and format normalisation
Values are rewritten to a single standard — g/m², cm, % — so every column can be compared, sorted and reused whatever the source wrote.
Controlled values and taxonomy
Finishes, compositions and constructions snap to an approved value list and land in the right attribute type and taxonomy branch, not free text.
Rule-based exception flagging
When a value breaks a defined rule — out of range, ambiguous, contradictory — it is flagged for review instead of passing through silently.
founder · semantic data · 0→1
