Wildcard DNS

*.imyme.app

Caddy

on-demand TLS

Next.js ISR

revalidate 1y

Cache handler

survives deploys

NestJS API

auth · billing · sites

BullMQ

resumable jobs

LLM router

primary → fallback

Piscina

FFmpeg pool

PostgreSQL

relational

MongoDB

flattened tree nodes

Redis

cache · OTP · limiter

R2

media objects

Built an LLM abstraction across AWS Bedrock (Nova) and Google Vertex AI (Gemini) with primary→fallback failover and per-model cooldown on quota exhaustion.

Enforced schema-valid output on Bedrock's Converse API — which exposes no native response_format — by constraining it to a JSON-schema-typed tool call.

Ran generation as resumable BullMQ jobs with stale-job recovery, and per-generation cost telemetry (tokens, latency, retries, USD against a typed price-per-model registry) surfaced in an admin dashboard.

Designed a polyglot persistence layer: PostgreSQL for relational data, MongoDB for a flattened tree-node model enabling diff-based partial updates instead of full-document rewrites on every builder action.

Implemented multi-level Redis caching covering domain lookup, ISR tag revalidation, OTP storage and a sliding-window rate limiter via Lua scripts.

Wrote a custom Next.js ISR cache handler that persists to disk and survives deployments.