You'll own the engine that turns prompts and references into usable training data: ComfyUI workflows, LoRA fine-tunes, and the detection / VLM models we benchmark them against. The work spans the full loop — generate, annotate, train, measure, improve.
You'll work directly with the founders on what gets shipped. There's no committee. If a workflow needs rewriting or a new detector needs training, you do it.
What you'll do
Design and maintain ComfyUI generation pipelines for security, defense and robotics scenarios.
Train and fine-tune LoRAs on domain data (weapons, drones, PPE, vehicles, environments).
Train and benchmark YOLO detectors and Vision Transformers on synthetic + real data.
Run synthetic-vs-real evaluations and improve the generation loop based on what fails.