Thnkrr
A professional network for high-school students, safe by design.
- Years
- 2026
- Type
- Professional
- Discipline
- Full Stack, Backend, AI
- Stack
- Typescript, Fastify, PostgreSQL, Drizzle, React, Vite, TanStack, AWS, AWS Bedrock
- Links
- Visit ↗
Thnkrr is an AI-matched mentorship, networking, and learning platform for students aged 13 to 18. As lead engineer through SingleBit, I took it from a working MVP to a platform rebuilt for scale.
Because the users are minors, compliance is the moat. COPPA, FERPA, SOPIPA, US state design codes, and India's DPDP shape the data model, the onboarding flow, and the moderation stack. Every decision is written down in a decision log with sources.
The API is Fastify + TypeScript on Postgres with Drizzle, with internal auth (argon2) and S3 media behind presigned URLs. The web app is a React 19 + Vite SPA on TanStack Router and Query.
What I did
- 01Rebuilt the platform end to end across 220+ commits: API, data model, SPA, and infrastructure
- 02Designed a staged moderation pipeline that runs before publication, with all models on Amazon Bedrock
- 03Built guardian-consent onboarding verified by schools, plus per-student privacy controls
- 04Benchmarked search and load at pilot scale before choosing production database sizes
- 05Wrote the architecture, safety runbook, and AWS deployment decision records
(Inside)
01
Pre-publication moderation
Every post, message, and profile field is checked before anyone else can see it. Our own rules screen first, then a policy model on Amazon Bedrock (retention locked off), then a human queue with named reviewers.
02
Guardian consent, verified by schools
Students under 18 name a guardian. The school confirms that guardian against its own records before a consent email is ever sent.
03
Roles and surfaces
Students, mentors, and educators get a feed, gamified learning, mentorship matching, appointments, messaging, and educator dashboards with PDF reports.
04
Sized before it scales
Search and load were benchmarked on a pilot-sized network (9,000 students, 700 mentors, 300 educators) to choose production infrastructure before the first school.