Overview
YourMitra is a personalized AI storybook platform — digital, B2C-event, and hardcover formats. I built it end to end as the sole engineer: backend, frontend, the AI generation pipeline, payments, deployment, and the admin panel that runs the business. It's live with real paying customers and repeat preview traffic.
The naive pipeline took ~20 minutes per story. Moving generation onto Temporal workers and parallelizing the independent steps brought it to ~2 minutes, with generation fully separated from delivery.
Key Features
Generation pipeline
- Child photo becomes an avatar, placed into a story template via a GPT image model, with text/overlays composited using Pillow.
- Dynamic (personalized) and static audio narration via ElevenLabs.
Platform engineering
- A single token-bucket rate limiter across OpenAI/ElevenLabs/Replicate, admin-tunable per provider.
- DB-driven feature flags, a 6-char short-link service, and a multi-provider notification engine (email, SMS, WhatsApp, RCS).
- Razorpay payments integrated end to end.
Tech Stack
- FastAPI
- PostgreSQL
- Redis
- Temporal.io
- Astro
- React
- OpenAI
- ElevenLabs
- Replicate
- Razorpay
- AWS S3
- Docker
Architecture
- FastAPI + PostgreSQL + Redis
- Temporal workers for parallel generation
- Docker services: API, worker, Temporal, Postgres, Redis
- Astro + React + TypeScript
- AWS S3 for media; Sentry + PostHog
- Self-managed VPS, domain, and reverse proxy
Results
~20 min
~2 min
Temporal parallel workers
Live with real paying customers