Anduril Lattice · Palantir AIP · Shield AI · 18 weeks
The 18-week path to AI engineering roles in defense. Modern deep learning + the domain knowledge (mission planning, multi-agent coord, classified-data handling, RAI policy) that separates "ML engineer" from "ML engineer on a defense product."
Modern deep learning + reinforcement learning fluency.
CNNs, transformers, training dynamics.
Skip this only if you've published a deep-learning paper. Otherwise, work through.
Q-learning, policy gradients, PPO. The math + practical training tricks.
Most defense AI for autonomy is RL-flavored. Spinning Up is the standard intro.
Custom datasets, distributed training, mixed precision, profiler.
You will write PyTorch every day. Be fluent past tutorial level.
Mission planning, multi-agent, autonomy stacks.
How autonomy is described in mission terms — go-to-waypoint, ISR loiter, contested overflight.
The vocabulary defense engineers use. Read DoD doctrine and recent autonomy papers.
The reference defense-AI platform — what's public is enough to learn the shape.
Read Anduril's engineering blog, public talks. Their architecture is the modern blueprint.
The other dominant defense data + AI platform.
Palantir publishes more than Anduril. Their docs + demos are worth a week.
Smaller, focused stack on autonomy for ISR + strike platforms.
Shield publishes papers (e.g. AlphaDogfight); reading them is the fastest way to understand their stack.
Quantization, TensorRT, federated, on-device inference.
Get a model from a 7B-param prototype to something that runs on a Jetson.
Most defense AI runs on size-weight-power constrained hardware. This is the engineering work.
NVIDIA's production inference path.
Same TensorRT as the perception roadmap; for defense it's table-stakes for any platform job.
Train across distributed nodes without centralizing data. Increasingly relevant for classified domains.
TFF and Flower are the open-source frameworks; concepts more important than tooling here.
Lattice OS, Palantir AIP, DoD Responsible AI Strategy.
The policy framework every defense AI engineer is expected to know.
Public document. Required reading; mentioned in many job descriptions.
Joint All-Domain Command and Control — the umbrella program for defense data integration.
Mentioned in every defense-AI interview. Know what it is and what it isn't.