Varun Daiya
I'm a CS and Physics student. Most of my time goes into building at the intersection of machine learning, quantum algorithms, and the mathematics that runs underneath both.
Currently building Terminus: a 2.6B-parameter native ternary (1.58-bit) small language model designed for offline, low-memory execution on consumer Android devices, supporting ten Indic languages.
I learn best by implementing things from scratch. Papers are a starting point and working code is where understanding actually lands.
I write occasionally on Substack.
// Selected Projects
NNCS-Mamba
Under DevelopmentBuilding a neural controller for drones. This project uses formal mathematical verification to ensure the AI-driven flight is safe and stable.
ampamp
SourceA Python library for quantum algorithms. It provides tools for "amplitude amplification", a core technique used in quantum search and optimization.
VectorCLM
SourceA custom-built AI language model with 145 million parameters. It was trained from scratch on 3.5 billion tokens to understand human language patterns.
MambaReview
Under DevelopmentA systematic study of different versions of the Mamba architecture. This project focuses on how these models process sequences of information differently.
WorldModels
SourceAn AI system that learns to "imagine" and predict its environment. It uses a combination of neural networks to build an internal model of the world.
// Publications
Congestion-Free Routing on Quantum Chips
arXivProposing swap-free spectral-qudit routing on quantum chips, where higher qudit levels act as orthogonal buses for nonlocal control delivery on near-neighbor hardware.
Quantum Amplitude Amplification Survey
A comprehensive survey of quantum amplitude-amplification methods and their algorithmic variants, bridging foundational concepts with modern optimizations.
// Open Questions
Are recursive AI machines the end-state, and was Karpathy's autoresearch the first primitive step toward them?
Are we moving toward an agentic software economy, and will we eventually see dedicated browsers and infrastructure designed exclusively for agents?
// AI Music
Generative sound and music experiments. Exploring how AI can interpret and create complex structural patterns in audio.
// Writing
An essay on the limits of Moore’s Law, the end of brute-force AI scaling, and the transition toward efficient intelligence. Exploring how one-bit models can bring powerful AI from massive data centers to everyday devices, changing how intelligence is built, accessed, and controlled.
Essay on computation, intelligence, and the ideas that connect them. I write about how information has evolved through history, why the shift from storing and retrieving to generating on demand is a paradigm shift, and what it means for technology, power, and the people building in this moment.