Ash Katnoria

ASH KATNORIA

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

AI | BI | CI | DI

Inspired by Bret Victor's explorable visualizations.

PAST

Started at a startup in 2000, then spent twelve years at ABN Amro and RBS building pricing and trade-capture systems for derivatives trading across Tokyo and Singapore.

Worked mostly on interest-rate derivatives and exotic products. Later helped set up an ION-based electronic trading platform for fixed income.

Pivoted to data science and machine learning in 2017. Now back in startup land - full circle.

PRESENT

I build things. Mostly software, for work, as well as for myself.

Right now I'm CTO at ThinkVAL in Singapore. It started with a no-code data platform for F&B and retail with advanced analytics and forecasting capabilities. But users struggled to pull insights from dashboards on their own.

When GPT-3 dropped, we built an AI chatbot on top of our data layer. As models got better, so did the chatbot. Now we're rolling out an agentic layer with sub-agents that write code, generate charts, run analyses, and answer how-to questions via RAG.

THINGS I'VE BUILT

DQN Tutorial (2018) Step-by-step Deep Q-Network guide for Lunar Lander. → Read

MDN Primer (2019) Mixture Density Networks explained. Most-read article on my site. → Read

CLIP Explorer (Jan 2021) Zero-shot vision-language search playground. Built when CLIP was fresh off the press. → Read

Face-Match (2018-19) - CLIP-based image embedding API and webapp. "Which celebrity do you look like?" 😄

Latent-Space Styler (2019) VAE tool for real-time face morphing. Drag vectors to add beards, sunglasses, hairstyles. → Demo

One-Shot DB Schema (2022) Prompt-engineering pipeline that turns an app idea into normalized Postgres schema with sample data in <30s. Early GPT-3 era experiment. → Demo

More on GitHub

FUTURE

Keep building, keep learning, stay honest. Work toward financial independence and the indie hacker dream