Research

Dependable intelligence in real systems

Two questions about what it takes for probabilistic AI to hold up in the real world.

Question 1

Dependability

How do probabilistic AI systems become trustworthy, observable, reproducible, and reliable?

determinism and consistency · replicability · verification and evaluation · observability · calibration · knowing when not to answer

Agents rarely fail because the model is weak; they fail because the system around it was never legible enough to verify.

Question 2

What changes when intelligence has to run under latency, privacy, cost, energy, offline, and physical constraints?

small and local language models · edge inference · privacy and data locality · latency, cost, energy · offline operation · cloud/edge routing

On a device without connectivity, the interesting part is the seams: what stays local, what escalates, and how a system knows the difference.

Where the two questions get tested.

Small models on real devices — The Acropolis Edge Lab's two projects investigate offline Hinglish voice Q&A and people counting on student laptops and Android phones, measuring quality, latency and hardware limits.
Physical and geospatial systems — Cameras, sensors, and imagery. A decade of maps taught me how unforgiving the physical world is about being wrong.
Agentic systems and operations automation — Agents expose the debt a system already carries: fragmented data, knowledge in people's heads, workflows that exist nowhere in machine-readable form. Operations automation at enterprise scale is where it shows first.

2026-08-26
Small AI And Small Talk
A week in Bangalore - take aways on Small LMs and enterprise AI - plus good conversations