Physical AI
Engineers moving from software-only AI to systems that act in the world, where being confidently wrong has consequences you cannot undo.
Length
7 weeks, part-time
Commitment
~6 hours per week
Cohort
12 live sessions, 20 seats
What this track assumes
Python. No robotics background and no hardware — the track runs entirely in simulation.
What you cover
7 modules
Each one is worked through live, then built on between sessions.
- 01Why embodied AI is a different problem — latency, irreversibility, and the missing undo
- 02Simulation first — building a world you can fail in cheaply, and where the simulator lies to you
- 03Perception under uncertainty — what the sensors give you and what they only appear to
- 04From policy to motion — control loops, planning, and the gap between the two
- 05Abstention in the physical world — a machine that stops when it does not know
- 06Sim-to-real — domain randomisation, calibration, and the transfer you cannot fully test
- 07Operating something that moves — safety envelopes, telemetry, and incident review
You build it, then you defend it.
There is no exam and no pass for attendance. Some people do not pass first time.
The project
A simulated robot of your own that completes a task and, more importantly, refuses it under conditions you defined in advance — with evidence of both behaviours.
Read first
Some of this is free
The knowledge library covers a good deal of the same ground. Start there, and pay us only once you have decided the reviewed version is worth it.
Evaluation Harnesses For LLM Systems
The piece almost every team skips. How to build a golden set, design a rubric, use LLM-as-judge without being fooled by it, and turn "is this better?" from an argument into a measurement.
Inference Cost As An Architectural Constraint
Model-backed features carry a variable cost that rises with success. Context budgeting, caching, routing and cascades, batching, and the unit economics you need before the architecture hardens around you.
RAG Architecture That Survives Production
Retrieval quality sets the ceiling on answer quality, and chunking decides retrieval quality. Hybrid search, reranking, grounding and citation, re-indexing, and how to tell which half of the system broke.
Testing Non-Deterministic Systems
Assertion-based testing assumes one correct output. Model-backed systems do not have one. What replaces equality checks, where deterministic tests still belong, and what the test pyramid becomes.
Other tracks
The rest of the technical family
AI Foundations
6 weeks, part-time
People who already use these tools daily and want to build with them — analysts, product people, and engineers early in their careers.
AI-Native Delivery
5 weeks, part-time
Teams adopting AI-assisted engineering who want the throughput gain to survive contact with review and production.
AI Engineering
8 weeks, part-time
Engineers building LLM applications, retrieval systems and agents who need them to hold up in production.
Agentic Systems Engineering
7 weeks, part-time
Engineers building systems that act on their own, and the leads accountable for what those systems do unsupervised.
Web3 & Smart Contract Security
10 weeks, part-time
Engineers moving into contract development, and teams preparing a protocol for audit and mainnet.
Mobile Delivery
5 weeks, part-time
Mobile teams whose release cadence is set by store review rather than by their own decisions.
Take Physical AI.
6 weeks, 12 live sessions, 20 seats, priced by region from $450. Places in the first cohorts are arranged directly.