multimodal AI · vision-language stack · hardware
ABOUT THE ROLE
Join an early-stage industrial AI startup as a founding engineer building the multimodal AI system for wearable assistants used in field operations. You will own key parts of the production vision-language stack, from model orchestration and evaluation to deployment on real-world hardware.
WHAT YOU'LL DO
- Build production agentic vision-language pipelines that support multi-step visual reasoning and tool use in industrial workflows.
- Develop evaluation frameworks, capture model failures, and create data and fine-tuning loops to improve model quality.
- Adapt model orchestration and edge inference to variable connectivity, latency, and deployment constraints.
- Create real-time voice and video interfaces, including conversational and proactive assistance features.
- Fine-tune and optimize open-source multimodal models for enterprise and on-premise deployments.
WHAT WE'RE LOOKING FOR
- Two to six years of experience shipping production vision-language or multimodal AI systems, with ownership of model and orchestration components.
- Strong Python and PyTorch skills, plus experience with tools such as vLLM, Triton, Ray Serve, ONNX, TensorRT, Hugging Face Transformers, Docker, and quantization methods.
- Applied experience with supervised fine-tuning, RLHF, evaluation design, model serving, and deployment; pretraining from scratch is not the focus.
- Experience building multi-step agents with real tool use, evaluation harnesses, and data improvement loops.
- A record of shipping AI products for end users and taking ownership of product o