Arnab Nandi ⭐️

Excited about a new thing: I built a management library you can talk to. 5+ years of coaching founders, noticed they all run into the same stuff. I kept using the same frameworks. So I turned them into cards you can read in five minutes or paste into an LLM that helps you actually implement them.

I launched my 40 AI Agents 6 months ago. Here's my productivity gain👇 Month 1 : 0.0% Month 2 : 0.00% Month 3 : 0.000% Month 4 : 0.0000% Month 5 : 0.00000% Month 6 : 0.000000%

@_philschmid I've been very impressed this feature since it's launch -- is there any documentation on how this works? (or did y'all just index all of YT?!)

One of the most underrated features of Gemini is that i can ace minutes/hour of video understanding in seconds! Below is an example of how to analyze Youtube Videos with a single API call using the Gemini Interactions API! Give it a try! You will be surprised how much progress we made.

@MarcJBrooker Define a word Define count Define text Define storage Answer will depend on definitions of these 4

Here's the technical interview question I'd actually ask. Suppose you have 5 MB of text, how would you count the words? What about 5 GB, TB, PB, EB? When would your approach change and why? What if you had to do it once, or once a week, day, hour, or second?

I've been waiting for something like this! Finally.

- SOTA model drops - solves a bunch of outstanding problems - everyone, "AI is magic no engineering needed" - AI gets stuck on next set of problems - everyone, "engineering is needed" <-- we are here - repeat

The year(s) of agents are going to spur a massive uptick in cybersecurity needs. Even worse, sufficiently complex SQL will be hard to ratify as safe. Who is working on Agentic SQL Safety?

This is so great!

Why can't we talk to our videos? LLMs and agents have drastically accelerated text-based interaction. For video however, performance gaps slow down the interactive loop. Excited to open-source Vidformer, our declarative optimization stack for video! ixlab.github.io/vidformer/

@rauchg @yikudoxyz :) Indeed. Unlocking natural language has created a temporary regression to text-centric. Excited for the hockey stick that's coming!

@yikudoxyz Notice I said natural language, which is all-encompassing

Terminals used to accept shell commands, now it’s natural language. IDEs used to accept code, now it’s natural language. Search engines used to accept keywords, now it’s natural language. Your sites & apps accept clicks, drags, hovers, taps… soon it’ll all be natural language

Maybe it’s time to bring back Perl as a mainstream vibecoding language? Fast and succinct; the biggest downside to Perl was that it was hard to read. But now that no human reads code any more, maybe it’s time to bring it back?

I don't think "agent memory" is capturing bucketing of context. Who is working mind palaces?

I am writing an article about elderly vibecoders If you are coding AI agents and are over 24 please DM me

@skalskip92 The tracker code shows the object detection / tracking, is the segmentation happening elsewhere?

track objects in real-time. use the trackers library. I work on this full time. the next release lands in February. link: github.com/roboflow/track…

winter olympics are coming. time to have some fun. RF-DETR + Trackers hockey demo and more coming soon: github.com/roboflow/sports

@basedsystems In the off chance academic venues are on your task list: hilda.io/2026/

@arnabdotorg We built a very large scale disaggregated compute and storage warehouse sort of like snowflake but more for biomedical / omic information. It’s also much faster than snowflake. Will write a post about it sometime

We built an ultra-fast genomics viewer for the browser. Before YouTube, sharing videos online was painful. Files were too large to email, too large to host, formats were not standardized, and each video existed in isolation instead of as part of a wider database. This is the situation we are in with omics and other medical data. Current viewers are either slow browser tools with few features, or desktop applications requiring large file downloads before use. We want to make complex biomedical data more accessible over the internet, and critical to that is a fast, secure genomics viewer in the browser. Today we're happy to show off our viewer. It performs like a desktop application, doesn't require large file transfers, and enables secure collaboration. We don't have open access yet, but if you'd like to give it a spin, shoot me a DM hox.bio/blog/genomics-…

@jamesvclements Nice!

@arnabdotorg Pretty quick

I built a tool that adds before and after screenshots to all my PRs

As the agentic internet grows, we're going to see a revisitation of the pre-WWW internet. The first one will be the reinvention of the "finger username@domain" protocol (1971, RFC 742), that showcased ".forward, .plan, .project and .pubkey from the user's home directory."