Key learnings
AWS ecosystem including Lambda functions, DynamoDB, and Polly for serverless video generation
Cloud cost management and optimization strategies for computational workloads, collaborating on shared infrastructure across a complex cloud environment
Video and speech processing pipelines with text-to-speech synthesis and audio-visual synchronization
Description
This cloud-based platform generates short-form videos in the style of popular TikTok content. Users provide text prompts and background video clips, and the pipeline converts text to speech via AWS Polly, synchronizes subtitles, and renders the final video file.
The application runs on AWS serverless infrastructure configured using SST. Lambda functions process speech synthesis and subtitle alignment, while DynamoDB tracks generation jobs.
Reflection
TikTok Money Glitch was a practical exploration of serverless architecture and audio-video processing pipelines on AWS.
Our main technical hurdle was managing execution timeouts and memory allocations in Lambda when processing audio synthesis and video rendering jobs. We optimized the workflow by offloading preview generation to client-side canvas rendering, invoking full Lambda video compilation only for final exports.
Building infrastructure via SST gave us hands-on experience structuring cloud stacks and managing resource limits in a team environment.
