AI/ML Engineering Intern: System 1 Models for Reactive Space Operations
Loftorbital - Toulouse, Occitanie - Global - Construction & Infrastructure
Wanna join the adventure?
Loft Orbital flies other people's payloads on standardized satellites, giving our customers a fast, simple, and reliable path to orbit. Our AI for Space team is looking for a motivated intern to explore how fast, lightweight “System 1” models can boost reactivity, performance, and usability across our satellite operational chains.
Reactivity use cases are at the heart of Loft's vision, and we are scaling them: from single-satellite reactive loops, to two-satellite / two-sensor tip & cue (a detection from one sensor automatically cueing an acquisition by another), and ultimately multi-satellite scenarios. Today we mostly rely on “System 2” models — powerful but slow and heavy at inference — from our partners. “System 1” models (fast, intuitive) are coming back strongly and could add value not only on reactivity, but also on performance and ease of use. This internship will assess that added value and prototype it on our reactivity scenarios.
Location: Toulouse office, city center (Jeanne d’Arc metro station) Duration: 6-month, end-of-studies internship — target start date: early 2027.
About this role:
Review the state of the art on System 1 / fast, intuitive models and compare them with the System 2 models used today, including our partners' models.
Map where System 1 models could bring value in our operational chain for reactivity use cases, from detection to tasking.
Define benchmark scenarios of increasing complexity with the team: a single-satellite reactive loop, then a two-satellite / two-sensor tip & cue loop, and, as a stretch goal, a multi-satellite scenario.
Set the latency budget, accuracy needs, and execution context for each scenario.
Build datasets and evaluation sets from Loft data, partner outputs, and open datasets.
Train or fine-tune System 1 models and benchmark them against System 2 baselines on reactivity (latency), performance (accuracy, compute footprint), and ease of use (integration and operational effort).
Prototype an end-to-end reactive loop, possibly as a hybrid design where System 1 triggers and System 2 validates, and assess how it extends to multi-satellite coordination.
Document results and recommendations for production use, and deliver a final demo.
Must Haves:
Final-year engineering or Master's student specialized in ML / deep learning / computer vision.
Solid Python and PyTorch skills, and good software engineering practices.