AI Engineer (Robotics/CV/Autonomous Driving background)

Equity + Competitive pay

We build physical AI agents for discrete manufacturing and warehousing. Our agents ingest multimodal physical signals — video, vibration, thermal, PLC telemetry — reason over structured operational knowledge, and drive closed-loop decisions autonomously.

We are looking for engineers who have worked seriously in robotics, computer vision, autonomous vehicles, or physical AI — and who have shipped production agentic systems.

Who you are

• Background in robotics, CV, autonomous vehicles or physical AI — you understand real sensor data and real-time systems from the inside
• Have built AI systems from 0 to production as an early engineer at a high-growth startup
• Strong instinct for tiered system design — you know when a threshold check beats an LLM call and when frontier reasoning is genuinely necessary
• Serious about production constraints from day one: inference cost, latency, failure modes, human oversight
• Understand the edge vs cloud tradeoff technically — model sizes, quantisation, memory footprints on constrained hardware
• Motivated by problems where the output changes something in the physical world

What you have done

• Shipped agentic systems in production with multi-step orchestration, tool use, state management, confidence scoring and human-in-the-loop gates
• Worked with physical-world signal data under real constraints — RTSP video, time-series telemetry, vibration FFT, thermal imaging, PLC / OPC-UA, noise, drift, latency, temporal alignment
• Designed tiered reasoning architectures combining deterministic rules, classical ML and LLM orchestration — knowing when each layer is appropriate and how to avoid over-relying on frontier models
• Built multimodal pipelines combining vision (YOLO, segmentation, action recognition) with structured operational context
• Worked with open-weight models including self-hosted inference, quantisation and context engineering for bounded reasoning tasks
• Thought seriously about edge deployment — model size vs capability tradeoffs, inference on Jetson-class or equivalent constrained hardware
• Integrated AI systems with ERP, CMMS, WMS or PLC layers across heterogeneous schemas

What you will build

• Tiered reasoning architecture across Vision Quality, Predictive Maintenance and Operations Planning agents — deterministic, classical ML, local LLM, frontier API — designed to minimise frontier dependency without sacrificing reliability
• Physical signal pipelines: video processing, YOLO, FFT feature extraction, cross-sensor correlation across vibration, thermal and current
• Agent harness: context assembly from ontology and memory, MCP tool dispatch, confidence scoring, approval gates, tracing, token budgets, swappable inference backends
• Local model serving on cloud GPU clusters and Jetson AGX Thor class edge hardware
• Industrial ontology mapping plants, assets, sensors, work orders and failure patterns — with a memory layer that compounds across deployments
• Operational surfaces: alert evidence bundles, replanning tools, audit trails