How do you get AI to seek the right data in the real world instead of drowning in all of it?

 In this episode, I sit down with Tara Javidi (UCSD professor and AI researcher) and Sam Bigdeli (repeat founder & former semiconductor supply‑chain exec), co-founders of Kav AI, to talk about “active physical intelligence”—hypothesis‑driven, curiosity‑led AI that hunts for the signals that matter in physical systems.

We cover:

• Why passive, data-soaks-everything AI hits a wall in the physical world
• Hypothesis-driven learning: letting models ask “what should I look at next?”
• From oil & gas spills to structural failures—predicting the next “leak” like a language model predicts the next word
• Handling massive, messy, multimodal sensor streams in real time (volume of context, not just length)
• Interpretability when your model is deciding which sensor to query and why
• What academia gets wrong (and right) about startups—and vice versa
• The hardest part of moving from novelty-driven research to problem-driven product
• How (and when) to disagree productively as co-founders

Links mentioned

• Website: www.kavai.com
• Company LinkedIn: https://www.linkedin.com/company/kav-artificial-intelligence/
• Sam Bigdeli: https://www.linkedin.com/in/sam-bigdeli-5310b923/
• Tara Javidi: https://www.linkedin.com/in/tara-javidi-28b450155/

🎙 Connect with me
LinkedIn – https://www.linkedin.com/in/greg-toroosian