In production
Synaply
AI-powered language learning platform
Solo-built production service running at synaply.me: REST API, background processing, PostgreSQL storage and Redis session state, deployed in Docker behind Caddy with automatic TLS and a full CI pipeline. 300+ commits since April 2026. Piloted with two classes (~45 students) at a Norwegian secondary school in spring 2026; a second pilot with the current version starts in September 2026.
- Authentication layer built from scratch — JWT access and refresh sessions, email verification, password reset and role-based middleware on top of chi.
- Core learning engine: vocabulary management and lesson flow combining FSRS v4 spaced repetition with a Redis-backed round-robin queue that recycles failed words within a session before scheduling them for long-term review.
- Anthropic Claude API integration for word analysis and exercise generation using tool use with strict JSON schemas, plus DeepL for translation — explanations and phonetic transcriptions in each learner's own alphabet.
- Layered architecture (handler → service → repository) covering 200+ REST endpoints documented with Swagger, on a PostgreSQL schema managed through 60+ versioned migrations; repository layer covered by integration tests against a real PostgreSQL instance in a container.
- Delivery pipeline: multi-stage Docker builds, GitHub Actions CI running build, vet, govulncheck, race-enabled unit tests and integration tests on ephemeral PostgreSQL/Redis containers, container releases to GHCR, and a Prometheus / Grafana / Loki stack with alerting.
GochiPostgreSQLpgxRedisDockerGitHub ActionsClaude APICaddyPrometheus
synaply.me Private repository — walkthrough on request