INDEPENDENT AI RESEARCH LAB • EST. 2024

Rethinking the Cost of Intelligence

Andril AI Lab is an independent research initiative explores computationally efficient approaches to predictive modeling, investigating both statistical methods and neural network architectures.

Explore Research About the Lab
Research Interests

Current Explorations

01. Infinite-Lag, Multi-Matrix MTD Models

MODEL TRAINING IN PROGRESS

Geometric-Scale Count Aggregation in Infinite-Lag, Multi-Matrix MTD Models for Next-Byte Prediction.

(Paper & Code will be published shortly)

Exploring geometric-scale count aggregation in infinite-lag, multi-matrix Mixture Transition Distribution (MTD) models for next-byte prediction as a computationally inexpensive alternative to neural network-based prediction in niche applications.

The work also explores distributed and modular training, where compatible models trained on different datasets can be combined by aggregating transition counts, supporting distributed, collaborative model development on consumer hardware with limited compute, memory, bandwidth.

02. Binary Neural Networks

PRELIMINARY RESEARCH

Exploring BitNet-based approaches to identify efficient neural network representations and investigate how far low-bit neural computation can go in terms of model performance and computational efficiency for niche applications.

Research Philosophy

Predictive AI Within Everyone's Reach.

Andril AI Lab explores ways to make predictive AI more accessible to independent researchers and practitioners across disciplines. The goal is to develop models that are affordable to train, practical to implement, and understandable enough to inspect, adapt, and build upon.

By investigating lightweight statistical methods, efficient neural networks, and distributed training approaches, the lab seeks to reduce dependence on the computational resources and infrastructure associated with frontier AI. The broader aim is to enable individuals and domain experts to develop predictive models for specialized problems that might otherwise remain underserved.

Leadership & Background

About Kathiravan

I hold a B.E. in Electronics and Communication Engineering and have spent over a decade working in web design, web application development, and mobile app development, alongside independent experiments in electronics and hardware.

My interest in AI research began around 2012 with independent explorations in natural language processing (NLP), using Princeton WordNet to investigate semantic relationships in language. I later collaborated with my friend Vivek on a separate exploration using a custom dataset. The work was eventually discontinued due to limited computational resources.

In 2024, advances in open-weight models and accessible machine-learning tools renewed my interest in AI research, leading me to establish Andril AI Lab. My current work explores statistical approaches to next-byte prediction and efficient neural network architectures, with a broader goal of making AI research and model development more accessible through computationally inexpensive methods.

Ventures & Timeline

  • Andril Web Technologies Web & App Design/Development Agency (Since 2014)
  • Andril AI Lab Independent AI Research Lab (Established 2024)
  • Bambaram Studios Game Development Studio (Upcoming 2027)

Contact

Reach out for research inquiries or technical collaborations:

[email protected]