Founder & CEO of Lossfunk, an AI research lab exploring foundational questions in AI and beyond. Previously founded Wingify (makers of VWO). I write at invertedpassion.com.
π Vibe Psychophysics Β 
A collection of classic perception and cognition experiments you can run in your browser. Each experiment includes background theory, an interactive paradigm, data collection, and results visualization.
π Murmuration
A Chrome extension that transforms your ChatGPT and Claude conversation topics into beautiful, animated black-and-white visualizations on every new tab.
βοΈ LongShot Β 
Get personalized screening recommendations for now and the future. Add them directly to your calendar.
π Explainers Β 
Interactive, visual explainers on topics like diffusion models, the Fourier Transform, biological scaling laws, cellular automata, and how LLMs work. Built for understanding through play.
π€ How LLMs work
A presentation I gave on how LLMs work.
An interactive exploration of all 256 elementary cellular automata rules. Visualize rule space structure, sweep initial conditions, and measure information propagation through spacetime.
πΉοΈ Atari from Pixels
Training a neural network to play Atari games directly from raw pixel inputs using deep reinforcement learning.
Detects logical contradictions in text using two approaches: embedding-based inconsistency vectors and LLM-as-Judge with topical clustering. Built with Sentence-BERT and GPT-4o-mini.
A browser extension that simultaneously searches on reddit, hacker news and other websites as you do Google searches.
Generating abstract art through neural networks in PyTorch. Feed in parameters, get unique art pieces out.
A single deep network that does four things: image search, image captioning, finding similar words, and finding similar images. Exploring the power of shared representations.
𧬠Deep Neuroevolution
Evolving deep neural network agents using Genetic Algorithms β no gradients required. A gradient-free approach to reinforcement learning.
Bayesian neural network implementation using Pyro and PyTorch on the MNIST dataset. Quantifying uncertainty in neural network predictions.
π¬ Follow me on X/Twitter for updates.



