Current Research
Foundations of Multi-Agent AI
I am working on modeling interacting AI systems in which information, objectives, communications, and observations are distributed across multiple agents.
In particular, working on mathematical descriptions of coordination, strategic behavior, information flow, agent interactions, and safety/performance trade-offs that arise when we apply a given mitigation.
The longer-term goal is not to develop a theory tied specifically to the present-day LLM architectures, but rather to understand which abstractions continue to make sense across different kinds of intelligent systems.
Information Flow and Protocol Design in LLM Multi-Agent Systems (2026–present)
Undergraduate Thesis / ongoing research
My recent work studies how information, communication constraints, and system behaviour interact in LLM-based multi-agent systems.
I began by analysing how much private information can remain recoverable when an external observer has only restricted access to the relationship between an agent’s internal state and its messages. I approached this using information-theoretic and probabilistic models together with controlled LLM experiments.
I am now extending this toward a broader protocol-design problem. Rather than assuming one fixed notion of safety, I am interested in frameworks where utility, communication, confidentiality, coordination, and environmental assumptions can be varied explicitly, so that the resulting trade-offs become mathematical objects we can reason about.
Publications & Manuscripts
Monitoring Limits and Hidden Information Flow in LLM Multi-Agent Systems (2026)
Information-theoretic and empirical analysis of hidden information flow and the limitations of passive observation in LLM-based multi-agent systems.
Manuscript in preparation, COSIC, KU Leuven
Cryptanalysis of the Legendre Pseudorandom Function over Extension Fields (2026)
[Solo author] Cryptanalysis of Legendre-PRF constructions over extension fields using differential and geometric-query techniques.
Side-Channel Based Attack Optimization and Reverse Engineering for ML-KEM Hardware (2026)
[First author] Side-channel analysis of an FPGA-based ML-KEM implementation, used to reverse-engineer the accelerator’s internal execution schedule and pipelining.
IEEE Computer Society Annual Symposium on VLSI (ISVLSI), 2026
Research Experience
Research Intern (June 2026 – August 2026)
COSIC, KU Leuven, Belgium
Worked on theoretical and empirical questions around information flow and safety in LLM-based multi-agent systems.
Developed an information-theoretic and probabilistic framework for studying hidden information flow under limited observation, together with controlled LLM experiments investigating how these effects appear in actual agent communication.
The work led to broader questions around communication constraints, coordination, system utility, and the extent to which safety properties can be guaranteed without unnecessarily restricting useful behaviour.
Returning to COSIC from November 2026 to February 2027 as a visiting scholar while continuing my undergraduate thesis research.
Undergraduate Student Researcher (Remote) (January 2026 – April 2026)
Bocconi University, Milan
Worked on a memory-conscious ARM Cortex-M4 implementation of the FAEST post-quantum signature scheme.
Supervisor: Prof. Emmanuela Orsini
Cryptography Research and Development (April 2025 – November 2025)
ECE Department, IIT Roorkee
Worked on software and FPGA implementations for fully homomorphic encryption, including CKKS-based encrypted computation.
Earlier Technical Experience
Undergraduate Research Assistant, Advanced Robotics Lab, IIT Roorkee (July 2024 – February 2025)
Worked on control systems and automation: literature review, proposed improvements to robotic control strategies, and multi-DOF robotic arm experiments.
E-Powertrain Engineer, IIT Roorkee Motorsports (March 2024 – February 2025)
Electrical division of the Formula Student team. Designed a DC to 3-phase AC inverter PCB for the electric drivetrain, and contributed to motor control strategies and battery interface design.
Dvole
Independent Research Initiative (2026–present) · dvole.org
Dvole is an independent research initiative I recently started for pursuing foundational questions that do not necessarily belong to one discipline.
A recurring idea behind it is to begin with some phenomenon, experiment, model, or collection of observations and ask what is actually structural about what we are seeing. Which parts depend on the particular representation, system, or experimental conditions? What survives when those things change? And can whatever remains be described more precisely or mathematically?
Some of the first work naturally grows out of questions I am already studying in AI and multi-agent systems, but Dvole is deliberately not an AI lab. I want it to be able to support serious work across mathematics, computation, physics, intelligent and complex systems, epistemology, and other foundational problems whenever there is a sufficiently interesting question connecting them.
It is still very early. The immediate goal is simply to develop a few serious research programmes and gradually build a body of mathematical, empirical, and technical work around them.
Selected Technical Work
Memory-Optimized FAEST v2.0 Implementation for ARM Cortex-M4 (Apr 2026)
A heap-free, memory-conscious embedded C implementation of the FAEST v2.0 post-quantum signature scheme for the ARM Cortex-M4.
FPGA Accelerated Convolutional Unit over CKKS Homomorphically Encrypted Data (Sep 2025)
Designed a custom hardware kernel on the Kria KV260’s FPGA fabric to accelerate convolution over CKKS homomorphically encrypted data, aimed at secure inference.
BFV Homomorphic Encryption Scheme Library (May 2025)
Implemented the BFV homomorphic encryption scheme from scratch in C++, including polynomial arithmetic, NTT-based operations, key generation, encryption/decryption, and homomorphic evaluation.
Research Interests
- Intelligent and Multi-Agent Systems: Coordination, interaction, strategic behaviour, information flow, safety, distributed decision-making, and protocol design.
- Mathematical Approaches to AI: Information theory, probability, game theory, dynamical systems, formal methods, formal epistemology, causal abstraction, and automata theory.
- Broader / Exploratory: Singular Learning Theory, Type Theory, HoTT, topology, mathematical physics, Algebraic QFT, and General Relativity. (Things I read and think about, not areas I claim expertise in.)
Selected Honours
Global Winner, AMD Open Hardware Competition (2025)
Adaptive Computation Track
Won the global Adaptive Computation Track for an FPGA-accelerated convolution unit operating over CKKS homomorphically encrypted data. Team entry (Aurva).
Supervisor: Dr. Tharun Kumar Reddy Bollu
2nd Place, CSAW Embedded Security Challenge (2025)
Solo participant
Attacked and mitigated firmware vulnerable to side-channel and fault-injection attacks on the ChipWhisperer Nano board, competing as a one-person team (FossilizedPluto).
Supervisor: Dr. Sparsh Mittal
Education
“I have never let my schooling interfere with my education.”
B.Tech in Engineering Physics (2023–2027)
Indian Institute of Technology Roorkee
Minor in Mathematics and Computing
Relevant Coursework: Linear Algebra, Calculus, Probability and Statistics, Mathematical Methods, Mathematical Physics, Complex Analysis, Number Theory, Graph Theory, Real Analysis, Topology, Functional Analysis, Quantum Mechanics, Statistical Mechanics, General Relativity.
I am not, as per the "definition," a good student. I have skipped almost all the lectures and tutorials, not taken assignment submissions seriously, and left exam halls as soon as I could. The reason is simply that the institute has unmotivated and uninspired academics (a few are good, but I am speaking on average), and their unnecessary demands.
But with my self-education, based on which I am literally building my entire research trajectory, I consider myself a very serious academic, where I sit down with a topic, for hours, to form abstract interpretations, writing long essays to get my raw thoughts structured and coherent, and also doing problems (so that I can justify to myself that I hate exams and assignments not because I can't solve problems. I can).
A little more about me
I started college thinking I would become a theoretical physicist. I later found the more abstract nature of mathematics a better fit for me. In the long term, I see myself pursuing pure mathematics full-time, but especially in my 20s, when my brain performance is at its peak, I want to absorb as much as I can and use that to apply my comfort with abstraction to as complex systems as possible.
Apart from that, I play guitar and drums (punk rock and funk), and absolutely love writing. I write things that are (I think) reasonably formal to half-baked and straight-up flawed arguments. I have also gotten into reading philosophy texts. I used to avoid them because I think nothing could be more harmful than reading others' philosophies without first developing a philosophy of your own through your own experiences. Once you have done that, do not accept their philosophies as the right ones, but only for comparison. You should get insights sometimes, but most of the time, it will just be interesting to see how different experiences other people have had.