AI tutoring startup
2026An AI tutoring and exam-preparation platform used by more than 13 million secondary-school students across web, iOS and Android. I replaced message-count entitlements with cost-based usage metering across every plan tier, built the AI tutor's memory pipeline, and added credit top-ups with the usage administration interface.
- Designed and shipped cost-based usage metering across every plan tier on a TypeScript, Next.js, tRPC and PostgreSQL platform, replacing message-count entitlements at millions of requests a month.
- Built the AI tutor's memory system as an asynchronous pipeline on Inngest durable execution: an LLM extraction pass mines durable facts out of conversations and a nightly reconciliation pass dedupes and prunes them, spread across a four-hour window so overnight database load stayed flat.
- Added credit top-ups alongside subscriptions and built the usage administration interface end to end, with a purchasing-power-parity Stripe price catalogue and Stripe as the source of truth.
- Worked in agile sprints directly with the CTO and the product team, and gave an external QA team several dozen tests covering every usage path.