About

We develop lean software and hardware to enable intelligent computing at one watt.

Our interests

generic causal prediction

We developed a generic approach to identify causal predictors first for a prediction target, then make probabilistic prediction based on the extracted set of them. One embodiment is Time Series Terminal https://tsterm.com. The engine behind can be configured with data from any domain: agriculture, climate, etc.

lean software

micrograd is a 20 kilobytes automatic differentiation library, found at the core of any machine learning framework. It is transparent and easy for kids to learn, for researchers to experiment with, and yet competitive in performance. It paves the way for simplifying attention-based models.

lean hardware

RISC-V is a family of open instruction sets for microprocessors. The V extension groups instructions of vector operations, backbone of AI calculations. There are open designs of hardware for RISC-V that run at about one watt.

References

Hannart, A., J. Pearl, F. E. L. Otto, P. Naveau, and M. Ghil, Causal counterfactual theory for the attribution of weather and climate-related events, Bulletin of the American Meteorological Society, 97(1):99-110, 2016.

Koenker, R. (2005). Quantile Regression (Econometric Society Monographs). Cambridge: Cambridge University Press. doi:10.1017/CBO9780511754098

Komunjer, Ivana, Quasi-maximum likelihood estimation for conditional quantiles, Journal of Econometrics, Volume 128, Issue 1, September 2005, Pages 137-164.

Diebold, F.X., Gunther, T. and Tay, A., Evaluating Density Forecasts, 1998.

Platform, method, and system for a search engine of time series data, United States patent US11893069B2.

Time Series Terminal, https://tsterm.com

TensorFlow, Apple’s MLX and our micrograd, /2025/09/25/tensorflow-mlx.html

Introduction to RISC-V, /2025/08/11/first-riscv64-program.html

The V extension for vector operations, /2025/10/31/rv64-v.html