RESEARCH

Research into the systems around intelligence.

We build experiments that make model behaviour, reasoning, interaction, and learning easier to inspect.

OUR THESIS

We study what happens after models become capable.

As AI systems act across time, capability depends on more than model weights. Tools, state, feedback, reasoning processes, environments, and failure recovery become part of intelligence itself.

We study the machinery around intelligence.
RESEARCH DIRECTIONS

What we investigate.

01

Agentic Systems

Long-horizon reasoning, tool use, stateful interaction, and autonomous behaviour.

02

Model Intelligence

Internal representations, mechanistic interpretability, model intervention, and behavioural analysis.

03

Reasoning & Verification

Search, symbolic constraints, structured reasoning, and verifiable computation.

04

Simulation & Learning

Reinforcement-learning environments, synthetic experience, robotics, and adaptive learning systems.

05

Benchmarking & Red Teaming

Agent benchmarks, reward-hacking analysis, adversarial evaluations, and repeatable tests for unsafe optimisation and deceptive behaviour.

Noma decision model graphic: state and typed questions pass through a fact channel, an 18-layer backbone and decision heads to a calibrated answer, with median latency against other decision models
FEATURED RESEARCH

Noma: an open, calibrated decision model

An open-weight decision model for agents and pipelines. It reads a state and typed questions and returns calibrated probabilities, an abstain signal, and an uncertainty estimate in one forward pass, in about 16 ms, with no generated text.

Paper: Noma: An Open System One Decision Model from a Sliced Decoder and Trained Heads. Atul Saxena and Divyanshi Sharma · Preprint, October 2026.

Inspect the systems behind the research.

Explore Projects