BreakTest

RoleFounding Engineer
Period2024 - 2025
Tech Stack
Next.jsDjangoPostgreSQLRedisStripeAI/MLRadix UITanStack Query
01

The Mission

BreakTest is an answer to a fairness problem.

THE STATUS QUO

Students pay thousands for private tutors or get lost in generic, system-less free resources. They practice randomly without knowing what's holding them back.

THE SOLUTION

I built BreakTest to provide the system—clarity, targeted feedback, and a path that adapts as they improve. It tracks mistakes, trains specific weaknesses, and measures real progress.

02

The Loop

Most score gains don't come from doing "more questions." They come from fixing repeatable errors.

01

Practice

Take a full test or targeted set.

02

Log

Every miss is logged and categorized.

03

Analyze

Mastery updates by error type.

04

Target

Platform generates specific redo sets.

05

Learn

Cognito (AI) gives progressive hints.

06

Improve

Score projections update in real-time.

That loop is the product.

03

What I Built

Full-length SAT testsRealistic timing and module structure.
Mistake VaultCaptures every error, tags it, and tracks mastery.
Cognito (AI Tutor)Progressive hints + real-time streaming.
Advanced AnalyticsBased on accuracy, pacing, and difficulty mix.
BillingSubscriptions + access control with Stripe.
04

Engineering

Auth Architecture

Clerk → Django integration. I implemented RS256 JWT verification in Django with custom middleware. This includes JIT user provisioning and performance-focused caching (Redis for tokens ~30s, JWKS ~1h). It enforces subscriptions via DRF mixins, effectively blocking unauthorized API calls even if the UI is bypassed.

High Performance

Built to scale. Uses PgBouncer for connection pooling in production. Implemented Redis-backed DRF throttling with per-endpoint classes for AI and submission endpoints. Heavily optimized DB queries using `select_related` and `prefetch_related` to eliminate N+1 query issues.

AI Guardrails

Designed to be helpful without leaking answers. Implemented a progressive hint ladder (Concept → Strategy → Technique) with context-aware prompting. The system prioritizes pedagogical value over simply giving the solution.

END OF FILE

Return