ML Runtime Engineer

Loft Orbital; Toulouse; Sur place; EUR 90 000 - 120 000; Plein temps

Avantages offerts par ce poste

Equity

35 days PTO

Health and life insurance

Lunch vouchers

Cross-office travel: SF, CO, Toulouse

Team off-sites

Relocation assistance to Toulouse

Résumé du poste

Loft Orbital is hiring an ML Runtime Engineer in Toulouse to turn partner ONNX models into hardware-optimized engines while balancing power and thermal budgets. You’ll work across graph compilation, runtime internals, and quantization, delivering end-to-end performance on flight hardware.

You’ll own benchmarking, profiling, and the budget validation process, collaborating with the Space Inference Engine, SDK, and the OBSW/Runtime team.

Qualifications

• Proficiency in C++ and Python for performance-focused development.
• Experience with model compilation for edge devices and accelerators.
• Strong knowledge of ONNX Runtime and related inference optimizations.
• Experience deploying ML on embedded GPUs (e.g., NVIDIA Jetson) and benchmarking.

Responsabilités

• Own model compilation: convert ONNX models into optimized engines per accelerator.
• Build the ground-side compile and delivery path, handling engine mismatch detection.
• Develop performance tooling: benchmarking, profiling, and budget validation.
• Contribute to Space Inference Engine, SDK, and collaborate with the Runtime team.

Connaissances

C++ & Python

TensorRT

ONNX Runtime

Embedded deployment

Perf tooling

CI/CD with HIL

Description du poste

Loft's AI for Space group is building the stack that runs real machine learning in orbit; turning raw sensor data into decisions on the satellite itself, so insight reaches the customer in minutes instead of hours. The AI Engineering team is where models meet the spacecraft. We own the full path from a partner's trained model to something that actually runs — fast — on flight hardware: compiling ONNX models into optimized engines for each accelerator, the Space Inference Engine and its execution providers, the SDK that packages models for flight, and the perf tooling that proves a model fits its power, thermal, and latency budget before it ever leaves the ground. As we open the platform to a growing roster of defense and Earth-observation partners, the team's scope and the range of silicon we target are expanding fast. As our ML Runtime Engineer, you'll own model compilation and the performance tooling behind it. We're looking for someone scrappy — the kind of engineer who takes an unfamiliar model and an unforgiving power budget, digs into the graph, and gets it running on real hardware without waiting for the path to be handed to them. You'll move between model compilation, runtime internals, quantization, and hardware-in-the-loop CI, sometimes in the same afternoon.

About this role:

• Own model compilation: turn partner ONNX models into optimized engines per accelerator (TensorRT, Hailo, AMD/ROCm), within power & thermal budgets
• Build the ground-side compile & delivery path; define how a mismatched engine is detected/rejected on-node
• Build performance tooling: benchmarking, profiling, operator-coverage matrices, and budget validation
• Contribute to Space Inference Engine (execution providers) and the SDK (model build & packaging); coordinate the OBSW / Runtime team

Must Haves:

• Strong C++ and Python
• Model compilation: TensorRT (and/or equivalent graph compilers)
• ONNX Runtime, quantization & inference perf optimization
• Embedded / edge GPU deployment (NVIDIA Jetson)
• Benchmarking & profiling / perf tooling
• CI/CD incl. hardware-in-the-loop

Nice to Haves:

• Hailo SDK and/or AMD ROCm compilation (iX10)
• Writing custom kernels / operator plugins
• Remote sensing / large-image handling
• RF / IQ signal data exposure
• Meson build, Yocto / minimal Debian images
• Model-weight protection / secure execution

Some of Our Awesome Benefits:

• Equity, we want you to have an active role in our success
• Up to 35 days of Paid Time Off (vacations & RTT ) and flexible working hours, we want you to be at your best
• Health and life insurance, we care about your health
• Lunch Vouchers, because let’s be honest, we love food! (we even have a slack channel about it #loft-gourmand)
• Cross-office travel opportunities between San Francisco, Colorado, and Toulouse to learn from our differences
• Company and team off-sites and many other events to work & celebrate together
• Relocation assistance to Toulouse when applicable
• Research shows that while men apply to jobs where they meet an average of 60% of the criteria, women and other underrepresented people tend to only apply when they meet 100% of the qualifications. At Loft, we value respectful debate and people who aren’t afraid to challenge assumptions.

Who We Are

Loft: Space Made Simple.

Founded in 2017, Loft provides governments, companies, and research institutions with a fast, reliable, and simple way to deploy missions in orbit. We integrate, launch, and operate spacecraft, offering end-to-end missions as a service across Earth observation, IoT connectivity, on-orbit AI, national security missions, and more. Leveraging our existing space infrastructure and an extensive inventory of satellite buses, Loft is reducing years-long integration and launch timelines to months. With more than 30 missions flown, Loft’s flight heritage and proven technologies enable customers to focus on their mission objectives. With a growing fleet on track to reach 30 satellites by 2027, we are scaling up quickly across our offices in San Francisco, CA | Golden, CO | and Toulouse, France to meet accelerating demand for space infrastructure.

We may use artificial intelligence (AI) tools to support parts of the hiring process, such as reviewing applications, analyzing resumes, or assessing responses and identifying potential inconsistencies or verification signals in application materials based on available information. These tools assist our recruitment team but do not replace human judgment. Final hiring decisions are ultimately made by humans. If you would like more information about how your data is processed, please contact us.