QCBRIEF
BosonQ Psi Federal will use a SpaceWERX SBIR slot to test a 2,000 parameter classifier against the ~20,000 unexplained orbital tracks the U.S. Space Surveillance Network logs each day.
Every day the U.S. Space Surveillance Network logs 18,000 to 25,000 observations it cannot yet match to a known satellite or piece of debris. Space-trackers call these routine blips Uncorrelated Tracks (Qubit Report). A New York startup called BosonQ Psi Federal just won its first U.S. contract to test whether a tiny classifier can sort them while running on satellite edge hardware, not in a data center.
The award is a SpaceWERX Open Topic SBIR, a Small Business Innovation Research slot that lets Space Force vet dual-use technologies outside its normal procurement cycle. BosonQ Psi Federal, the federal affiliate of BosonQ Psi Corp, says it will adapt its BQPhy quantum-inspired platform into a Physics-Constrained Quantum-Assisted Machine Learning classifier (PRNewswire).
The performance numbers are the company's own. BosonQ Psi says the model shrinks from roughly 14 million parameters to about 2,000, a 99% reduction, while retaining more than 99% classification accuracy, cutting inference latency up to tenfold, and drawing roughly 90% less power (Interesting Engineering). The "quantum-assisted" label refers to physics-constrained classical compute that runs on an NVIDIA Jetson Nano, not execution on quantum hardware.
BosonQ Psi Federal previously demonstrated an earlier version at the BMC3I TAP Lab (formerly SDA TAP Lab). The Space Force has not published independent benchmarks, and the contract value was not disclosed.