Sarvik Contact

/03

Research

Three threads I keep pulling on. Less about publications, more about where I spend my deep-work hours, and what I'm trying to make faster or sharper.

// active research

VLA-DSS

A compact vision-language-action model for robot manipulation

One question drives it: how much capability survives when you shrink a generalist robot policy by an order of magnitude? At 28.9M parameters, RGB-only, it runs in real time on weak hardware, and still holds its own against models several times its size.

  • Wavelet-scattering observation encoder, a Lipschitz-stable front end with provable robustness to small perturbations.
  • Frozen DINOv3 vision and FiLM language fusion, strong semantic features without the training cost.
  • A Fourier Neural Operator action head, band-limited and resolution-invariant action chunks instead of an MLP or diffusion head.
  • LIBERO Object 79.5%, competitive with Octo-Small at matched size and far smaller than SmolVLA (450M).
Read the full write-up See the reading list

The wider threads

01

Machine learning & robotics

Efficient vision-language-action models and robot learning: how a policy should represent and generate actions, and how to do it at a fraction of the usual parameter count.

// current objectives

  • Flow-matching action heads for the Octo generalist robot policy.
  • A city-scale traffic digital twin with graph neural networks and reinforcement learning.
  • Parameter-efficient policies: keeping capability while shrinking the model an order of magnitude.
02

Algorithms & systems

Advanced data structures and optimization for performance-critical systems, low-level design where every clock cycle counts.

// current objectives

  • Hardware-accelerated engine architecture: sprite batching, dynamic VBOs, ECS.
  • Competitive programming at the Expert level (Codeforces).
  • Geometric optimization solvers, e.g. a dual-phase MVEE word guesser.
03

Mathematics & quant

Mathematical modelling and optimization, including quantitative finance: statistical tools that discover and capture market anomalies.

// current objectives

  • Developing and backtesting Alphas within WorldQuant's quantitative frameworks.
  • Applying statistical analysis to find non-obvious correlations in high-dimensional data.
  • Optimization and applied mathematics across systems and ML.