Alexandre Carlhammar

Leading research on Rapidly Reconfigurable Manufacturing Cells for low-volume production. Deploying SOTA end-to-end RL policies, VLA models, and diffusion policy-based controllers to enable adaptive, sample-efficient, and high-precision smooth manipulation.

Founding engineer. End-to-end: concept through flight test. Ballistic launch, 300+ km/h sustained flight.

Built: Full-stack GNC for GNSS-denied autonomous flight and precision engagement. Multi-sensor fusion (IMU, baro, RGB/IR cameras). RL-based control modules. Airframe, wing structures, propulsion validation via CFD and bench testing.

Closed weekly build-test-fly loops in Mojave. Co-built the founding team. Co-led early fundraising.

Entered with zero defense background. Conducted 120+ interviews across government, industry, and research to isolate actual technical constraints. Iterated via prototypes as probes, presenting rough systems to experts to find what was wrong, missing, or unrealistic.

Awarded $80K by the Defense Innovation Unit to build and deploy a prototype.

AI models that internalize orbital physics and multi-satellite coordination for autonomous operations. Hyper-realistic simulations, open-source propagators, VAEs for physically consistent scenario generation.

Path planning for rover at lunar south pole (Blue Origin mission, 2026). Sun-synchronous traverses via SLAM. NeRF and 3D Gaussian Splats trained on lunar DEMs and orbit imagery. Real-time planning in hardware-in-the-loop simulation.

Won NASA's Lunar Autonomy Challenge. [Stanford announcement]

Large-scale multi-agent architectures for pharmaceutical automation. Advanced RAG with semantic chunking, query rewriting, multi-agent orchestration.

Led 7 engineers. Deployed system into Pfizer's production environment.

Problem: Deploying software to satellites required custom integration per mission. No standard runtime, no shared storage, no portability.

Built: OS-level abstraction for satellites—containerized app deployment across constellations, S3-compatible distributed storage, data relay through intermediary satellites, sensor virtualization. Essentially Kubernetes for space.

Deployed on ISS (Nov 2024). Upcoming NASA mission (Oct 2025). Grants from Swedish Space Agency and NASA.

Problem: Multi-satellite mission scheduling is NP-hard. Traditional solvers take weeks per iteration.

Self-taught deep RL over winter break. Convinced Airbus to share their high-fidelity orbital simulator. Trained approximate policies that generate feasible distributions in seconds.

Cut planning time from weeks to hours. Led first undergraduate workshop at AMLD 2024 (50+ attendees). [recap]