Solo, end to end
Typeza
A production AI SaaS built alone, from database to App Store.
Problem
Most AI features die between the demo and production. Building a full AI SaaS solo means owning the whole chain, not just the model call: the async pipeline, the billing, the mobile app, the infrastructure that survives a deploy.
Approach
Each analysis fans out 15 LLM calls at once through a Celery and Redis pipeline, then a consensus step weighs whatever returns instead of failing on a slow model. Real-time transcription runs over WebSockets. React web, native SwiftUI iOS, Stripe and Apple IAP, all on AWS ECS managed with Terraform.
15
parallel LLM calls per analysis
< 3 min
end-to-end processing
4
ECS services, 100% Terraform
DjangoCeleryRedisOpenRouterReactSwiftUIAWS ECSTerraform
Architecture
React / iOS
client
Django API
DRF
enqueue
Celery + Redis
task queue
fan-out
LLM call x15
OpenRouter
15 parallel calls, < 3 min
weigh
Consensus
partial-tolerant
Report
structured
WebSocket
live transcript
Stripe + Apple IAP
billing
AWS ECS x4 + Terraform
infra
Lesson
The model call is one line of code. Orchestration, partial-failure handling and billing are the actual product.
