Heston Labs is an applied ML studio building hard, unglamorous machine learning systems across markets, hardware and wherever else the problem turns out to be interesting. We started with Latent, a real-time embedding system that compresses 28 financial assets into a single 128-dimensional space. Less a product than a foundation for other things. From there came CARL, a benchmark for evaluating cross-asset models that didn't really exist before we needed it. We're also poking at a small hardware analytics device, which has nothing to do with markets.