Zamna Space builds autonomy software for satellites. Constellations are getting larger and cheaper while ground teams stay the same size, and more of what a spacecraft does has to be decided onboard rather than on the ground. We're moving applied AI onto the spacecraft the way it already has in robotics: detecting and recovering from faults, managing spectrum and adapting links, coordinating across a fleet.
The role
In September 2026 we begin a NASA SBIR Phase I award (EXPAND.3.S26B, NASA Ames) on autonomous onboard health management for small spacecraft and distributed systems. The work covers how a satellite recognizes that something has gone wrong, how it decides what the cause is with limited compute and no ground contact, and how a fleet of them reaches a shared conclusion over a link that is slow and often unavailable.
You will be the second technical member of the company and one of two named key personnel on the award. The areas you could have an impact on include, but are not limited to:
- Building the onboard software that decides what a spacecraft does when something goes wrong, from real-time detection through diagnosis and recovery.
- Applying modern ML to spacecraft telemetry, including model development, evaluation against real mission data, and the compute and power budgets that constrain what can run in orbit.
- Coordinating behavior across a fleet, where vehicles have to reach the right answer over links that are slow, intermittent, and often unavailable.
- Building the simulation and fault-injection tooling that everything else is tested against, and the evaluation campaigns that show it works.
Our engineers own the full lifecycle of what they build, including the experiment design, the evaluation, and the technical writing that goes to NASA.
Responsibilities
- Train, tune, and evaluate anomaly-detection models on multi-channel telemetry in PyTorch, and benchmark them against published real-mission telemetry datasets.
- Define the health-summary format vehicles exchange, and implement the messaging layer that moves it across constrained inter-satellite links.
- Build fault-injection scenarios and telemetry generation on top of an astrodynamics simulator, and maintain the fault library the evaluation runs against.
- Profile model runtime, memory, and power against representative flight-compute targets, on hardware or emulation.
- Integrate the pieces into an end-to-end system, run the evaluation campaigns, and write up the results.
Basic qualifications
- Bachelor's degree in computer science, engineering, math, or science discipline; OR 3+ years of professional software development experience in lieu of a degree.
- 3+ years building and shipping machine learning systems in production, or a graduate degree plus 2+ years.
- Strong Python; experience with PyTorch or TensorFlow on real data, not just notebooks.
- Fluent with AI coding tools such as Claude Code or Codex: fast output, healthy skepticism about what comes back, and the test design to catch what's wrong before it lands.
Preferred skills and experience
- Experience applying models to sensor, telemetry, or other time-series data from physical systems.
- Experience with distributed multi-vehicle or fleet systems operating under partial observability and unreliable communications.
- Working knowledge of distributed coordination and consensus, and the failure modes they hide.
- Detection of rare or unlabeled events in operational data.
- Edge or embedded inference: quantization, runtime and memory profiling, deployment to constrained compute.
- Robotics, UAV, automotive autonomy, or spacecraft experience; familiarity with state estimation, control, or fault management.
- C or C++ alongside Python; real-time Linux; flight software frameworks such as F´ or cFS.
- Experience with LLM tool-use or agent frameworks, particularly under constrained action spaces.
- Peer-reviewed publications, or the kind of experimental rigor that survives a NASA technical review.
- Excellent written communication. A meaningful part of the deliverable is prose.
Additional requirements
- Full-time, starting September 30, 2026.
- Hybrid in New York City: three days a week at South Park Commons, two days remote.
- Selection is subject to approval by the NASA Technical Monitor, which requires submitting your résumé and role as part of a key-personnel package.
- Two-person team, first federal contract, no existing process to inherit. You should want that.
Compensation and benefits
$165,000 salary plus equity and benefits.
Work authorization
Under the terms of the SBIR program, all research and development under this award must be performed in the United States. Applicants must be authorized to work in the U.S.