Seven disciplines. One practice of careful work.

A new paradigm for neural architecture — Reverse Synthetic Neural Networks that construct models directly from data patterns, without gradient descent, backpropagation, or GPU training. From zero-training synthesis to interpretable, efficient intelligence.

Edge-resident, link-survivable, auditable AI for the sovereign defence of the realm — recognition, anomaly and network-structure analysis that run on the device itself, offline, and explain why. Built on the RSN line so a model can be re-tuned in the field with no GPU and no cloud, and every call is one a human can audit. It is the research substrate beneath अभेद्य (Abhedya).

Hardware and algorithms in which randomness is not noise to suppress but the substrate itself — for ultra-low-power inference and neuromorphic systems.

Market microstructure, microdynamics, and the theory of financial markets as programmable computational systems — for asset management, risk inference, and the science of economic computation.

Cryptographic provenance for intellectual property of any kind. On-chain timestamping today only proves that one exact file existed at a moment in time — it breaks the instant a work is paraphrased, reformatted, translated, or remixed by a model, and it forces you to reveal the work in order to register it. We work on the unsolved version: provenance that is invariant to semantic-preserving transformation, and that can be proven in zero knowledge — letting a creator register a work and later establish priority or trace derivation without ever revealing the work itself. It is the research substrate beneath ठप्पा (Ṭhappā).

Statistical inference over sparse, noisy, high-dimensional observations of the universe — and, at the lab's leading edge, a single standard of evidence for exoplanet atmospheres. We build machine-accelerated Bayesian retrieval that turns the public JWST archive into reproducible, preregistered verdicts on contested biosignature claims. It is the research substrate beneath व्योम (Vyom).

The control intelligence for new-age robotics — built, not trained. We use Reverse Synthetic Neural Networks to compile a model for each robotic capability directly from human motion, with no GPU training: a library of drop-in skills — grasp and hand articulation, bipedal gait and balance, gaze and facial/mouth actuation — that run on the robot itself, in real time, at the edge. A small team can own a capability end-to-end, and each capability is independently and commercially valuable.

Modeling the machinery of living systems at scale — protein dynamics, regulatory networks, and the interpretable biology of disease.

Algorithms and error-mitigation techniques for the NISQ era, and the theoretical frontier of what quantum advantage actually means outside cryptography.