AI in Healthcare | Startup, NBA, Cricket | prev @techstars, @beondeck, @Cisco
Joined March 2009
- Tweets 9,124
- Following 2,971
- Followers 1,788
- Likes 7,803
419 Photos and videos
Bargava retweeted
🚨 New Workshop Alert!
Making a chart with AI is easy. Making a system that finds stories, works with new data, matches your editorial taste is much harder.
Join @karthiks at #VizChitra2026 for a hands-on workshop on agentic dataviz workflows.
More: vizchitra.com/2026/sessions/…
ALT VizChitra 2026 workshop announcement featuring Karthik Shashidhar. The session, “Building Agentic Dataviz Systems,” focuses on creating repeatable AI-assisted data visualization workflows rather than one-off chart-generation demos. The description highlights topics including building context packs, guiding AI models toward stronger hypotheses, generating and critiquing visualizations, and designing workflows that remain useful as data changes. The graphic includes links to learn more about the session and purchase tickets for VizChitra 2026. A QR code directs viewers to the event website. The workshop will be held as part of VizChitra 2026 at the Bangalore International Centre on 4 July 2026.
1
1
61
Great post and good argument on needing more companies to eventually get good sovereign AI.
Also, from the post - TIL: Involution.
1
84
💯
As knowledge workers, getting upskilled and getting good - both in the techniques and tools - is our version of "Did you sharpen your axe today?".
Jun 11
I can't tell you how much money I've spent on LLMs in the past 2 years - non trivial amount, let's say far more than anything I've spent on any of my hobbies in my lifetime combined; and whatever hope I have for my future is because I see myself as a one person business who invests money to earn money ; and taking to LLMs being comfortable with their outputs is a fundamental skill that needs you to spend tokens & constantly sample all those experiences to build an intuition around wielding it
1
62
For most applications, search is all you need. Good post on how @LangChain uses full text search for agent observability.
Jun 10
How do you support full-text search JSON filtering over agent traces that span up to hundreds of MBs, while keeping a median (P50) latency of 400ms?
Here’s an inside look at how we built a custom inverted index from scratch for SmithDB.
langchain.com/blog/full-text…
1
101
Bargava retweeted
Jun 10
Killer playbook for multi-Mac agent setup. Great "how to" guide for Agent Cookie to have your primary Mac and your MacMini agent machine be logged into all the same things with magic cookie sync. github.com/mvanhorn/agentcoo…
8
18
294
60,002
Bargava retweeted
Jun 10
This is too boring to go viral but it will save you someday.
11
19
262
82,848
Bargava retweeted
Jun 10
After interviewing for Research Scientist roles at DeepMind, Isomorphic, Meta, Cohere and more, I wrote up everything I learned. Technical prep, logistics, negotiation, and emotional breakdowns. Check out my guide: silviasapora.github.io/blog/…
46
300
3,634
232,551
A good post on where and how to find value.
If everyone else is just building a wrapper on top of frontier models and could be taken over by the frontier labs in one of their upcoming updates - how to build a startup that is resilient to this?
1
80
Is someone working on this for JEE/NEET prep product?
Jun 8
I poured my 10 years of teaching experience into a skill.
It's called /teach, and it can teach you anything.
Here's how it taught me to solve a Rubik's cube:
1
70
If you watched @srush_nlp's brilliant explanation on on-policy distillation and want to read more on it - the post by Li is a good read 👇
Jun 9
New blog post: On-Policy Distillation — Promise, Pitfalls, and Prospects.
OPD combines on-policy rollouts with dense teacher supervision.
But it is not a free lunch.
I discuss three failure modes and introduce our new paper.
louieworth.github.io/blog/op…
1
183
A very good and a very approachable post on text optimization.
2
70
And here you go - a good overview of loop engineering 👇
132
I got different answers from LLMs and also based on the reasoning effort - answers varied.
One used just one case (basically omit K Hearts). Another did two case. And xhigh versions of the models did three cases.
Here's a combinations question - that I am getting different answers from the LLMs.
What is your answer?
From a standard deck of 52 cards, how many 5-card selections contain exactly 2 kings and exactly 3 hearts?
29
At this point - I no longer use AGENTS.md - but rather use a markdown that orchestrates a bunch of skills.
For the relevant flow/feature that am working on - I call the appropriate skills.
1
112
Now: Harness Engineering
3 months: Goal Engineering
6 months: Fleet Engineering
Jun 7
Don’t worry it’ll take 3 months until it’s there.
We’ll be talking about fleets that design your loops then.
40
We are going to see lot and lot more of SLMs solving narrow tasks, but matching frontier model's performance.
Great model here 👇
Jun 8
Introducing 𝗖𝗔𝗦𝗦𝗜𝗡𝗜-𝟭.𝟬. A fully homegrown, Pareto Frontier opensource model for language-agnostic agentic context management from India. Our 4B parameter SLM achieves SoTA performance at 1/18th the cost. Link in the reply.
1
4
394
these are some good questions if you want to interview someone in RL 👇
1
88