JOB-1736
A seed-stage startup is putting AI co-pilots onto smart glasses for technicians working in server facilities, electrical utilities and aerospace. It needs a founding engineer to own the multimodal AI core from the beginning, from vision-language pipelines to real-time voice and video running on devices out on site.
Your focus
- Ship a production pipeline where a vision-language model reasons over multiple steps and uses tools, applied to real industrial procedures and inspections
- Build evaluation harnesses, a system for logging failures, and a loop that turns customer data into fine-tuning runs so each release measurably improves
- Handle edge inference and connectivity changes, including orchestration, runtime tuning and graceful fallback when latency or bandwidth varies
- Design hands-free voice and video experiences for the glasses, with natural dialogue and unprompted warnings tailored to different users
- Adapt open-source multimodal models using SFT, RLHF and quantization so they can run inside customers’ own facilities
About you
- Roughly two to six years delivering multimodal and vision-language products to actual users, using Python and PyTorch
- Experience owning both the model and the orchestration layer, not just demos or pure research
- Practical agentic work with real tool use, beyond RAG wrappers or single-turn chatbots
- Startup or founding-engineer background, or time at a similarly intense company
- A BS or MS in computer science, machine learning or engineering
- Authentic enthusiasm for vision, wearable devices and industrial AI
On-site five days a week in the San Francisco Bay Area. Visa transfers are considered.
