How much does it cost to train a state-of-the-art foundational LLM? $4M. Facebook's 65B LLaMA trained for 21 days on 2048 Nvidia A100 GPUs. At $3.93/hr on GCP, that's a total of ~$4M. Google's 540B PaLM was trained on 6144 v4 TPUs for 1200hrs. At $3.22/hr is a total of ~$27M

When you start a company, there's this illusion that at some point things will get easier. It never gets easier. You earn the right to solve problems with higher stakes. But it never gets easier. And if it starts to feel easy...you're probably missing something.

I am excited to share that @KodeziHQ has been ranked the #1 product on Product Hunt for the month of February! Thank you to everyone who has supported us along the way. There’s still much more to accomplish, but we can’t wait to see what the future holds for Kodezi!

Believe in your vision, even when the world doesn’t. In 2021, before the terms “AI” and “ChatGPT” were known to the mainstream, I was a teenager showcasing my startup Kodezi, at an event in Orlando. I remember demoing my AI debugging solution to investors and attendees, but I

Twitter just locked themselves out of their own API endpoints. Amazing

seems interesting

if i start going off, the openai board should go after me for the full value of my shares

Reflecting on a transformative 2023, my journey from ML medical research to building @KodeziHQ has been incredibly fulfilling. The privilege of connecting with a diverse group of people—from Founders and CEOs to Influencers, Investors, and even Actors—has truly enriched the

Aisle seats are more likely to be taken than window seats after Alaska Airline incident.

Someone asked me if HBO’s Silicon Valley was accurate and I said “no it is much weirder”—he thought I was joking until I told him about a talented VC friend who believes in Flat Earth yet still invests in space tech because “the government has to make it look like it works.”

Join us at #apidays Helsinki & North 2024 for "OpenAPI: Where It's Been and Where It's Going"! Ishraq Khan, CEO & Founder of @KodeziHQ, explores the evolution and future of OpenAPI. #apidays #OpenAPI #Innovation More info: apidays.global/helsinki_and_n…

I don’t expect engineers to say “thank you.” I actually prefer that they don’t. Last week, Sam Altman mentioned that people saying “please” and “thank you” to ChatGPT is costing OpenAI tens of millions of dollars in energy, because those extra tokens take up more compute. Wild

8 years ago, I accidentally built my first startup. At 13, I launched TeachMeCode — a peer-to-peer coding platform where kids taught kids. No teachers. No pressure. Just real learning. Thousands joined. Middle school me pitched investors. In 2019, it became Kodezi. Then: Help

To every founder waiting for the perfect launch date: Ship it in half the time. You don’t need to be 100% ready. You need to be live. Our early Kodezi features barely worked, but they were real. Real gets feedback. Real creates momentum. Waiting feels safe. But the market

Kodezi Chronos: A Debugging-First Language Model for Repository-Scale, Memory-Driven Code Understanding. arxiv.org/abs/2507.12482

For the past 3 years we have been working on building out our own versions of a debugging LLM at @KodeziHQ and our approach on how we would tackle the biggest issue for AI code models. We call Kodezi Chronos, a new way to approach code debugging. arxiv.org/pdf/2507.12482

Somewhere in the world, an unknown young person is probably starting a company now that will eventually be as big as today's tech giants.