Work · Client · the firm · 2026
Physics-informed AI for R&D
A platform helping a manufacturer's R&D teams move from need to concept, grounded in physics-informed surrogate models and a knowledge graph of prior art.
What was built
- Surrogate models via Latin Hypercube sampling and physics-informed networks
- A knowledge graph over patents and prior art
- A staged need → opportunity → concept gate process
What happened
The architecture answered the 'why not just an LLM' question well enough to earn outside validation. Physics-AI was a new domain for me; I learnt it while running the programme.
What it taught me
Defensibility against a plain LLM wrapper is answered with an architecture, not a pitch.