← Selected workAI discovery · App Store intelligence

Keyword Graph Explorer

A graph-based keyword discovery service that combines live App Store metrics, semantic expansion, and cross-run novelty memory.

RoleProduct architecture & implementation
VisibilityPrivate source
StatusWorking private service
WalkthroughAvailable on request
Challenge

What needed to change

Flat synonym lists do not explain whether a keyword direction is commercially viable, coherent, or exhausted. The service needed to explore multiple branches, validate them with real metrics, and report partial results honestly when a target could not be reached.

Approach

How the system works

Seed Mode grows depth-prioritized branches from a known keyword. Hunt Mode begins from a natural-language product context, proposes candidate seeds, validates them, and then launches the same graph exploration pipeline.

Candidate generation mixes semantic neighbors, pivots, mechanical variants, and escape strategies. Live metrics, duplicate detection, cycle prevention, and domain-drift checks decide which paths continue.

Engineering decisions

Trade-offs made explicit

Completion before vanity score.

A coherent branch that reaches the requested depth is more useful than a short high-scoring dead end.

Honest partial fulfillment.

The API reports shortfalls and reasons instead of fabricating a complete answer.

Memory with decay.

Useful leaves are rewarded across runs while exhausted directions lose priority over time.

Technology

Stack

  • TypeScript
  • Fastify
  • OpenAI
  • Model Context Protocol
  • SQLite
  • Zod
  • Vitest
  • React
Private product walkthrough

Guided demo available on request.

The walkthrough will cover Seed Mode, Hunt Mode, branch inspection, fulfillment reporting, and how live metrics influence exploration.