My own SaaS: a queue-driven pipeline that writes, assembles and renders short-form video at scale using LLMs, CLIP-based footage selection and serverless FFmpeg.
ClipNest automates the creation and distribution of short-form video content. A user describes a channel; the platform researches, writes, narrates, assembles and publishes videos on a schedule.
It is the hardest kind of system to build well: long-running, resource-hungry, failure-prone at every stage, and expensive if you get the orchestration wrong.
A microservices-inspired architecture orchestrates a multi-stage LLM scripting engine, CLIP-based footage selection and serverless FFmpeg rendering through a BullMQ job pipeline. Every stage is independently retryable and independently scalable.
Rendering runs on ephemeral compute so cost tracks usage instead of capacity. Infrastructure is defined in Terraform, so the whole stack is reproducible from an empty AWS account.
Intake
Research
Script
Voice
Footage
Render
Publish
Research, outline, script and hook generation as separate prompted stages with validation between them.
Embeddings match each script beat to the most relevant stock footage automatically.
FFmpeg on Lambda — parallel renders without paying for idle GPU-grade instances.
Terraform-defined ECS, Lambda, S3, queues and networking — no click-ops.
Everything I offer as "AI automation" on client projects — content pipelines, LLM-backed features, background processing at scale — I have already shipped and operated in ClipNest. That includes the unglamorous parts: retries, cost control, rate limits and observability.
Tell me what your business does and what you need. I'll come back with a scope, a timeline and a price.
Start a project