Ash Katnoria

ASH KATNORIA

EPISODE: 4, LR: 1e-5, EPS: 0.5

Inspired by Bret Victor's explorable visualizations.

past

I started at a startup in 2000. Then came twelve years at ABN Amro and RBS, in Tokyo and Singapore, building pricing and trade-capture systems for interest-rate derivatives and exotics.

Then life threw a curveball. The chart above shows roughly.

So I picked up something new: 3D Printing. it lead to small DIY projects such as controlling aircon, tv from arduino, Nest like DIY thermostat. And eventually to small robots but controlling robot using simple L|R, Drive|Stop wasn't good enough. I wanted to program it using something smarter, something that learns from feedback - enter deep RL. One thing lead to another and introduced me to the world of machine learning and artificial intelligence.

So I went from MIT's Analytics Edge in 2016, through Udacity Nanodegrees, Coursera specialisations, and Stanford's NLP, graph ML and reinforcement learning courses, to Stanford's Professional AI Programme in 2022.

present

Right now I'm tecnical Co-founder and CTO at ThinkVAL in Singapore. We started as a no-code data platform for F&B and retail: analytics, forecasting, dashboards. It turns out people don't want dashboards. They want answers.

So when GPT-3 landed, we put a chatbot on top of the data. The models got better, and so did the bot. Today we have a team of specialist agents that write code, draw charts, run analyses, answer how-to questions and

On the side, I occassionaly write here: long posts, short zines, and the odd interactive thing.

things I've built

  • 2026 Zines. Short, illustrated takes on building software in the AI era. → Read
  • 2024 One-Shot DB Schema. An app idea goes in. A normalized Postgres schema with sample data comes out, in under 30 seconds. → Demo
  • 2021 CLIP Explorer. Zero-shot image search, built the month CLIP came out. → Read
  • 2021 Face-Match. "Which celebrity do you look like?" An image-embedding API and web app. It went viral.
  • 2019 Latent-Space Styler. Real-time face morphing with a VAE. Drag a vector, add a beard. → Demo
  • 2019 MDN Primer. Mixture Density Networks, explained. Still the most-read post here. → Read
  • 2018 DQN Tutorial. A Deep Q-Network that learns to land on the moon, step by step. → Read

More on GitHub.

next

Work on things that buy autonomy, mastery and impact. Write down what I learn, failures included, so the next person gets there a little faster