Lead AI Product Engineer, Vision & Multimodal Systems

Good Glitch takes one client per industry, finds what their operation is leaking, and builds the AI module that closes it, self-hosted, on infrastructure the client controls. No retainers, no vendor lock-in. This role owns the visual and multimodal side of that work: products built on vision, voice, and data, not just text.

WHAT YOU’LL OWN

A visual AI product end to end, for one client at a time. You’ll sit with the people who actually run the operation before you design anything, because the signal worth building on is usually not the one everyone assumes. You’ll decide, module by module, what’s a classical computer vision problem (detection, tracking, segmentation, OCR, anomaly detection) and what’s a vision-language model problem (grounded description, visual question answering, multimodal retrieval, agentic visual reasoning), then build the smallest thing that proves it in production. Everything ships self-hosted. When it’s done, you hand over the code, the docs, and the reasoning, cleanly enough that the client never needs to call us back for that module.

WHAT WE LOOK FOR

Real, production computer vision experience that predates and doesn’t depend on VLMs: detection, tracking, segmentation, traditional ML on visual data, systems you can point to and explain the tradeoffs of. Alongside that, current, hands-on experience with vision-language models: fine-tuning, evaluation, and deploying them in real products rather than demos. The judgment to know which tool the problem actually calls for, and the discipline not to reach for an LLM out of habit. Comfort owning a visual system end to end: data pipeline, model, interface, all of it self-hosted.

HOW WE WORK

Small team, high ownership, no layers between you and the client. Text before visuals: decisions get made in writing before anything gets designed. Every engagement ends in a clean exit: full IP transfer, no dependency built in on purpose. You’ll work across verticals over time rather than being boxed into one industry forever.