Senior Full-Stack AI Systems Engineer — Dockerized LTX-2.3 GPU Inference Platform

Dockerized LTX-2.3 Inference Pipeline — No ComfyUI Runtime

I’m looking for an experienced ML infrastructure engineer to build a fully containerized LTX-2.3 inference pipeline.

This is not a traditional frontend-focused full-stack role. The ideal candidate is a strong software engineer with deep experience in Python, PyTorch, Docker, CUDA, cloud GPU deployment, model serving and production-grade backend systems.

I do not want ComfyUI to be used as the runtime interface. The final system should run directly through Docker and allow inference jobs to be submitted programmatically or through a simple command-line/job specification workflow.

The scope includes:

- Containerizing the complete LTX-2.3 inference environment with Docker.
- Managing model downloads, caching and persistent model storage.
- Supporting configurable inference parameters such as prompt, seed, resolution, duration, frame rate, output path and optional image/video conditioning.
- Accepting jobs through a simple and documented mechanism, such as JSON job files, CLI commands or a lightweight queue.
- Saving generated videos and job metadata to persistent storage.
- Providing clear logs, job status and useful error messages.
- Ensuring the container can run reliably on cloud GPU providers.
- Documenting deployment, configuration and usage.
- Avoiding ComfyUI as a production dependency or runtime interface.

Before implementation, I expect a short architecture proposal covering:

1. The proposed components and data flow.
2. How inference jobs will be submitted and processed.
3. Docker image structure and dependency management.
4. Model storage and caching strategy.
5. GPU and VRAM management.
6. Error handling, retries and job recovery.
7. Output storage and metadata.
8. How the system could later scale to multiple GPU workers.
9. Security and isolation considerations.
10. Estimated implementation stages and deliverables.

The first version should prioritize simplicity, reliability and reproducibility. I am not looking for an unnecessarily complex Kubernetes or microservices setup unless there is a clear technical reason for it.

Application instructions

Do not send a generic or templated proposal.

To be considered, your response must include:

- A brief description of a similar GPU inference or ML deployment you have built.
- Your recommended job-submission mechanism for this project and why.
- A high-level architecture for the solution.
- The main technical risk you expect when containerizing LTX-2.3.
- How you would handle model weights so they are not downloaded on every container start.
- The cloud GPU provider or deployment environment you would recommend.
- A realistic breakdown of the implementation into milestones.

Begin your proposal with the sentence:

“The main operational risk in this project is…”

Proposals that only repeat the job description, contain generic claims or do not answer the technical questions above will not be considered.

Expected deliverables

- Dockerfile and supporting configuration.
- Reproducible build and deployment instructions.
- Job submission script or CLI.
- Inference worker implementation.
- Example job configurations.
- Persistent model and output storage configuration.
- Logging and failure-handling implementation.
- Technical documentation.
- A working demonstration on a cloud GPU instance.