I'm an independent researcher working on provably stable control for the power grid. My work sits between control theory and neural-network verification: building learned controllers whose stability is guaranteed by a formal proof over a region, not inferred from testing, and making those proofs scale to grids large enough to matter. I'm about to begin my undergraduate studies and will be continuing this line of research there.
Most of my current work is in collaboration with
Prof. Wenqi Cui's group at NYU Tandon ECE, where I'm extending her published work on neural Lyapunov stability with formal certification, alongside power-systems modeling and grid-infrastructure research on the strain that large-scale compute places on distribution. Several of these threads are now written up and under review.
Research collaborator, Cui Group, NYU Tandon ECE · multiple manuscripts under review
OpenReview ·
GitHub
The grid line of work, and the projects nearest to it. Every project links to its repository; several are the codebases behind the manuscripts listed further down.
Instead of repairing a learned stability certificate after the fact, the network is trained to be verifiable by construction, by differentiating through the CROWN verification bound and making the size of the certified region part of the training objective. On the lossy 39-bus swing model it certifies a larger region than the prior counterexample-guided method (ρ=2.5 vs 1.8) at the same rate, verified three independent ways with zero soundness contradictions, and scales cleanly from a 2-D to a 4-D slice. The open question is how far it climbs toward the full grid state before the verification cost walls out.
A learned Lyapunov function can satisfy its stability condition on 99.98% of sampled states and still fail a formal verifier, because a proof quantifies over an entire region while sampling only ever visits a measure-zero subset. This work shows that gap concretely, 200,000 random samples find no violation where a directed search finds them 6.1% of the time, and closes it with counterexample-guided verification cross-checked three ways (CROWN, Jacobian bound propagation, and dReal SMT). It extends Cui and Zhang's 2022 work on the exact open problem their paper names.
A slack-ratio phase transition that decides when certificate descent on a learned system component is safe to trust, packaged as a rank-fidelity diagnostic with verified multi-domain results. Basis for a manuscript under review.
Formally-certified input-to-state-stable small-gain guarantees for heterogeneous swarms with learned dynamics, so a stability proof holds across agents whose models were learned rather than derived.
Recalibration-invariant fingerprints for industrial telemetry — the relational “atoms” of multi-physics operation, stable across sensor recalibration.
Power-systems modeling — grid infrastructure
ongoing
Build and run power-flow simulations for grid infrastructure design, gathering and analyzing the simulation data and producing engineering reports that inform decisions such as line selection and routing. Work conducted under confidentiality, described here at the level of capability rather than client detail.
Research Intern — data-center grid infrastructure
2025–Present
Researching power-distribution bottlenecks and grid strain driven by the exponential compute demand of large-scale data centers, including generation-and-tie planning for how new load interconnects to the grid. Role and findings held under confidentiality.
5+ manuscripts in review across Nature Computational Science, IEEE Control & Dynamics, and Nature Machine Intelligence. Several of the projects above are the codebases behind these; others are being written up now.
Azaad AI — a foundational model of Sikhi private repo
in development
An internal AI model developed at the Institute of Azaadist Studies that aims to understand Sikhi and its nuances at a foundational level — not surface retrieval, but the capacity to reason from first principles and hold genuinely foundational viewpoints. The project is in discussion with the Oxford AI Society regarding funding.