AwesomeGene
A fully automated video production pipeline: one brief in, a finished, published video out, with no human in the middle.
The problem
Video production is a sequence of slow, expensive, largely mechanical steps. Generating clips, stitching them into a coherent long-form edit, synthesising narration, timing subtitles, cutting vertical shorts, and publishing to a platform are each individually automatable — but chaining them is where it falls apart. Every stage is long-running, every stage can fail, and a naive script that runs them in sequence will lose an hour of GPU work to a single transient API error.
What we built
- 01
A staged pipeline where each step is an independently queued task with its own retry policy, so a failure at stage four does not discard the output of stages one through three.
- 02
A task queue and broker handling long-running generation work, decoupled from any request lifecycle.
- 03
Object storage as the handoff medium between stages, with job state tracked separately in a document store so any run can be inspected or resumed.
- 04
Containerised workers, letting the compute-heavy stages scale independently of the orchestration layer.
- 05
Publishing as its own guarded stage, so output is reviewable before anything goes live.
Decisions worth noting
Failure isolation between stages
Stage boundaries are durable checkpoints. Expensive generated artefacts land in object storage before the next stage begins, which turns a failed run into a resumable one instead of a total loss.
State separate from artefacts
Job state lives in a document store while media lives in object storage. Inspecting or reasoning about a run never means moving gigabytes of video.
Built for unattended operation
The pipeline is designed to run without supervision — which means every failure mode has a defined behaviour rather than an exception that stops the world.
More work
VetBid
An AI business-development platform that reads federal solicitations and tells veteran-owned firms which ones they can actually win.
Next Door Notice
A geospatial monitoring platform that watches every planning application in London and tells people when one lands near them.
Olim Paveway
An AI planning tool that turns a fragmented, high-stakes immigration process into a personalised, document-backed plan.
Tell us what is slowing you down
A short conversation is usually enough to tell whether this is a build, an automation, or something you should not do at all. We will tell you which.