One prompt in, a narrated and captioned 9:16 short out, on AWS Lambda.

Aegasis Labs, an AI-assisted software consultancy in London, wanted short vertical video made from a single request: a prompt in, a finished short out.

That one prompt can become a finished short, with every stage free to fail and retry without paying for the stages before it again.

Five stages, each one safe to retry, orchestrated on AWS Lambda: script, scenes, narration, captions, and assembly.

1. 01

   Script & scenes

   script(prompt) → scenes

   The prompt becomes a script, and the script becomes illustrated scenes, each a stage of its own.

2. 02

   Narration & captions

   narrate() · caption()

   Narration generated for the script, with captions timed word by word and rendered with PIL.

3. 03

   Assembly

   assemble(9:16)

   MoviePy cuts it together: Ken Burns motion on the stills, crossfades between scenes, all in a 9:16 frame.

4. 04

   Cost-aware retries

   s3.cached() or call()

   Every stage's output cached in S3, so a retry never calls a paid endpoint twice.

Alongside it, for an energy-infrastructure cybersecurity product, Hassan ported the production streaming chat adapter from PHP to Python, on OpenAI streaming.