Artificial Intelligence Research

Experiments at
the edge of AI

Vasuki Labs is an independent AI research collective. We build, break, and publish experiments across language models, reasoning systems, and multimodal intelligence.

See experiments Have a Conversation
0
Papers published
0
Active experiments
0K
Model downloads
Neural pathway active
↑ 94.1% accuracy
Epoch 847 / 1000
Loss: 0.0023
Experiments

What we're working on

View all →
Language
Chain-of-Thought Faithfulness
Measuring whether LLM reasoning traces actually reflect internal computations, or are post-hoc rationalizations.
Active · NeurIPS 2024
↗
Multimodal
Vision–Language Grounding
Probing how vision-language models form shared representations across modalities without explicit alignment supervision.
Active · ICLR 2025
↗
Reasoning
Compositional Generalisation
Testing whether transformers can recombine learned concepts to solve novel problems not seen during training.
In review · ICML 2025
↗
Live Demo

Run a simulation

Interactive visualisations of our core research. All computation runs client-side in your browser.

Simulation
Neural Network
Attention Flow
Token Embeddings
Layers 4
Neurons / layer 6
Activation speed 1.0×

A forward-pass visualisation through a fully-connected network. Signals propagate layer by layer; node brightness reflects activation magnitude.

Team

The researchers

Adamya R. Khairwal
Founder · Research Director
Executive, Managing Director, Technical HOD
Shaurya Chauhan
Design & Development HOD
Design, Deployment & Advisory Board
H
Hanuveer Sisothiya
Research Engineer | OS Advisor
Research, Development & Deployment
Stay in the loop

Join Vasuki Labs

Get notified when we publish new experiments, papers, and open model releases. No noise — just research.