← Projects

Work In Progress

Knowledge Graph of Taiwanese Tea Genetic Heritage

A public, browsable knowledge graph reconciling Taiwanese tea cultivar genetics — scattered across peer-reviewed molecular studies, institutional breeding records, and conflicting vernacular naming — into one queryable structure.

Purpose

Why this exists

Tea in Taiwan spans three genetically and historically distinct lineages: the wild endemic species Camellia formosensis, introduced/naturalized Assamica-type "ancestor trees," and cultivars bred since 1903 by the Tea Research and Extension Station (TRES). The underlying genetics are well-studied, but the data is fragmented — inconsistent identifiers across molecular studies, unstructured institutional web pages, and the same cultivar going by multiple non-interchangeable names and codes. Nobody has reconciled it into a single, internally consistent, machine-queryable structure. That reconciliation — not the tea itself — is the informatics contribution.

What it will do

Interactive map

Named tea-growing regions across Taiwan, browsable with no genetics or graph-database background required.

Lineage explorer

A searchable, visual view of how today's commercial cultivars trace back to wild or historically introduced ancestors — e.g. TTES No. 18 ("Ruby Red"), a literal genetic bridge between the wild endemic species and an Assamica variety from Burma.

How it's being built

Data & entity resolution

Cultivar and species records are extracted from primary literature and institutional sources, then resolved into a canonical entity model — recognizing when records using different identifiers refer to the same real-world cultivar — with full source provenance retained on every claim.

Graph model

Resolved entities are loaded into a Neo4j property graph, queried with Cypher, so questions like "which currently cultivated teas trace back to the wild endemic species" can be answered directly through graph traversal instead of a different table-join query for every possible path.

Public web app first

Built MVP-first: a small, genuinely working product covering a few cultivars and regions ships early, then grows incrementally — rather than waiting for "complete" data in a domain that has no natural stopping point.

Stack

Neo4jCypherPythonpandasDuckDBEntity Resolution

Status

Early-stage and actively evolving. Literature capture and note synthesis are underway; the Neo4j graph and public web app aren't built yet — but the reasoning, sources, and workflow are all public now.