Automated Short Form Video Creation

(Project)

I built this feature for <a href="https://tessact.ai" target="_blank" rel="noopener noreferrer">Tessact</a>. I designed a microservices system with four parts: an AI service (FastAPI + PydanticAI) that manages LLM calls, a core backend (Django) with Celery and RabbitMQ for job queues, a frontend (React + TypeScript) for uploads and previews, and a rendering service (Remotion on AWS Lambda) for video generation. Each service runs in Docker and deploys to GCP Cloud Run or AWS Lambda. I chose microservices so face tracking could scale on GPUs while other services stayed on standard instances. I tested LLM providers and switched from GPT-5 to Gemini, cutting costs by 60%. I tested transcription vendors and switched to ElevenLabs, cutting costs by 72%. The system turns 2-hour podcasts into 30-40 clips. Editing time dropped from 4 weeks to 1 hour. Cost per podcast dropped from $1,000 to $5. Output quality is 95% ready-to-post based on client feedback.