Embedded AI Engineer

Prophesee; Paris; Sur place; EUR 70 000 - 110 000; Plein temps

Résumé du poste

Prophesee in Paris is seeking an Embedded AI engineer to design, adapt and optimize perception models for edge deployment on constrained hardware. You will bridge AI, software and embedded engineering to meet latency, memory and power targets.

You will develop C++ and Python tooling for benchmarking and deployment, profile pipelines, and translate models for runs on edge accelerators such as Jetson or CUDA-enabled devices.

Qualifications

• Embedded AI / model optimization with edge-centric workloads.
• Strong hands-on ML/DL optimization for constrained targets.
• Experience with PyTorch, TensorFlow and optimization workflows.
• Ability to profile latency, memory and power.
• Experience with C++/Python tooling for deployment.

Responsabilités

• Design, adapt and optimize AI models for embedded and edge deployment.
• Profile end-to-end inference pipelines for bottlenecks.
• Convert models for target runtimes and accelerators while preserving accuracy.
• Implement and evaluate quantization, pruning and other optimizations.
• Develop C++ and Python tooling for benchmarking, profiling and deployment.
• Contribute to software architecture around memory management and real-time constraints.
• Establish reproducible benchmarks and compare trade-offs across targets.
• Support deployment on edge AI targets.

Connaissances

Embedded AI

C++

Python

Performance profiling

Inference optimization

Embedded Linux

ARM

Git

Computer vision

Outils

PyTorch

TensorFlow

CUDA

TensorRT

NVIDIA Jetson

Description du poste

Why PROPHESEE?

Inspired by human vision, PROPHESEE's technology combines patented sensors and AI algorithms to go beyond the limits of conventional approaches. It captures information differently, revealing dynamics that remain invisible to traditional systems. This breakthrough redefines computer vision and opens exciting technical challenges in cutting-edge fields such as autonomous vehicles, industrial automation, robotics, IoT, security and surveillance.

You will work where AI algorithms meet real computing constraints, helping PROPHESEE turn advanced perception into efficient edge implementations. The role offers the opportunity to optimize models and complete inference pipelines for demanding embedded targets where latency, memory, bandwidth and power consumption are as important as algorithmic accuracy.

Mission

Design, integrate and optimize AI / perception algorithms and models for deployment on constrained computing targets. The role sits between AI, software and embedded engineering and focuses on translating algorithmic performance into efficient production implementations that meet latency, compute, memory and power constraints.

Role description

• Design, adapt and optimize AI, deep learning and computer vision models for embedded and edge deployment.
• Profile end-to-end inference pipelines to identify compute, memory, latency and bandwidth bottlenecks.
• Convert or adapt models for target runtimes and accelerators while preserving required accuracy.
• Implement and evaluate quantization, pruning, graph optimization, kernel optimization and other model-compression techniques where relevant.
• Optimize pre-processing, post-processing and data movement around AI inference, not only the neural network itself.
• Develop C++ and Python tooling for benchmarking, profiling, conversion, validation and deployment.
• Contribute to software architecture and development, specifically around memory management, scheduling and real-time constraints.
• Establish reproducible benchmarks and compare accuracy / latency / power / memory trade-offs across targets.
• Support deployment on edge AI targets.

Key deliverables

• Optimized AI / perception models and inference pipelines for selected embedded targets.
• Reproducible benchmark suite covering accuracy, latency, throughput, memory and, where available, power consumption.
• Model-conversion and optimization flows for the selected inference runtimes and accelerators.
• C++ and Python tools for profiling, validation, benchmarking and deployment.
• Optimization reports documenting bottlenecks, trade-offs and improvements versus baseline implementations.
• Deployment-ready integration with the software stack and associated technical documentation.

Performance indicators

• Achievement of agreed accuracy, latency, throughput, memory and power targets on selected hardware.
• Performance improvement versus baseline model / pipeline implementations.
• Successful model conversion and deployment across selected target runtimes.
• Reproducibility and stability of benchmarking and validation results.
• Absence of unacceptable functional or accuracy regressions introduced by optimization.
• Maintainability and reuse of optimization tooling and deployment workflows.

Profile

• Embedded AI / model optimization — strong requirement — Hands-on experience optimizing machine learning or deep learning workloads for edge or constrained targets.
• Deep learning frameworks — strong requirement — Strong practical knowledge of frameworks such as PyTorch, TensorFlow or equivalent.
• C++ and Python — strong requirement — Ability to implement performance-sensitive inference code and supporting optimization / benchmarking tools.
• Performance profiling — strong requirement — Ability to analyze latency, throughput, memory use and algorithmic complexity and identify bottlenecks.
• Inference optimization — strong requirement — Experience with quantization, model conversion, graph optimization, kernel/runtime optimization or equivalent techniques.
• Embedded Linux / ARM — required — Understanding of embedded Linux, ARM platforms, cross-compilation and constrained computing environments.
• NVIDIA Jetson / TensorRT / CUDA — highly desirable — Hands-on deployment or optimization experience on NVIDIA edge AI platforms.
• Computer vision / perception — desirable — Experience with detection, tracking, image processing, event-based vision or related perception pipelines.
• Hardware-aware mindset — required — Ability to reason about CPU/GPU/NPU capabilities, memory hierarchy, data movement, power and target constraints.
• Benchmarking & validation — required — Ability to build reproducible performance benchmarks while tracking model accuracy and functional regressions.
• Software engineering practices — required — Git, testing, documentation, code review and maintainable production implementation.

Language skills

Fluency in French and English

Soft skills

Strong problem-solving skills, strong analytical skills. Flexible to dynamic environments and fast changing technologies. Passionate about technology. This person must work well with other engineers in a team environment. Good sense of autonomy. Must be pragmatic and self-motivated to complete a task even if it is outside of just the “well known” realm. “Can Do Attitude” is preferred.