Generative AI Pipeline Engineer (Tech Lead)

Role Overview

CapsLock is building an AI Media Production platform and is hiring a Generative AI Pipeline Engineer (Tech Lead) to architect and run it. The platform generates reproducible visuals of real products (1:1 match including hardware and geometry) with no hallucinations—a control and reproducibility problem, not a prompting problem.

Responsibilities

• Architect and run the AI Media Production platform, including:
  • Serving ComfyUI (or similar engines) and designing GPU environments
  • Building one-click UIs so non-technical teams can operate workflows
• Ship modular, versioned, reproducible generative pipelines (node-based and/or code-first, e.g. Diffusers)
• Create one-click presets and end-to-end use cases for image and video generation
• Train LoRAs end-to-end using real product photography, including model evaluation and ensuring production readiness
• Implement control-first generation across modalities using:
  • ControlNet, IP-Adapter, inpainting, and reference conditioning
• Engineer QC inside the pipeline with automated checks against product references before human review
• Build a creative intelligence loop to encode what makes top assets perform well into reusable presets/styles
• Own the technical roadmap with the Design Director and qualify new models/tools
• Drive adoption via enablement (training/enablement), not mandates

Must-Haves

• Shipped generative pipelines that are adopted and used daily by others
• Production-grade workflows using ComfyUI (or similar) or equivalent code-first experience (e.g. Diffusers)
• Strong Python for automation, API integrations, and custom nodes
• LoRA training end-to-end, from dataset curation through evaluation
• Control methods across modalities (t2i, i2i, t2v, v2v) including:
  • ControlNet, IP-Adapter, inpainting, reference conditioning
• Experience deploying pipelines as internal tools
• Enough visual literacy to set and enforce quality standards
• Confident English (B2+)

Nice-to-Haves

• 3D/render-hybrid workflows using Blender or CAD renders as geometric ground truth
• LLM orchestration/agentic workflows (e.g., automated prompt generation, VLM-based QC)
• Experience producing visuals for performance marketing

What’s Offered

• Fully remote environment
• Generous paid vacation, unlimited sick days
• Ongoing learning (books, workshops, conferences)
• Home office equipment support (cut off in the provided text)