Nirant
@NirantK
Followers
15K
Following
46K
Media
1K
Statuses
10K
AI engineer that takes companies from RAGs to riches
localhost
Joined December 2012
A stellar engineer and CTO friend is working on RL gyms/environments for browser and would love to speak to folks using browser agents. Who should he speak to?
3
2
2
What the heck @lovable_dev — I cannot access my own app data? This is just security theatre to create cloud lock I was on track to spend $5000 on Lovable. That is going to drop to $0 as I spend all my energy moving away
4
1
16
This Cybersecurity Awareness Month, there is a bill in Congress that actually threatens user security in major app stores while restricting consumer choice and punishing American success in digital markets, dubbed the "App Store Freedom Act." 🧵
1
3
9
I'm convinced coding CLIs are AGI: I've never gone deep enough into AWS Lambda to memorize its commands, some of it still baffles me! Yet, with CLI tools, I set secrets, re-ordered pipeline DAG steps, and tweaked configs—all in under 6 hours.
1
0
9
This is resolved now. Indian cards work. had already spent $100 on Lovable by then though, so I am unable to switch!
Can someone tag Emergent folks here? I can't use my US credit card and the Indian card does not work. I want to give them money and they won't accept it.
0
0
0
VC ecosystem is the most inclusive when it comes to quasi-technical events If you can't get that invite, you are doing something wrong Our GenAI group has a higher bar for a WhatsApp group ffs
0
0
2
Wrote 500+ lines of SQL slop and Claude restructured it cleanly into 4 queries inferring my intent 95% of the way LLMs are amazing.
0
0
3
Most popular models used for Hindi at OpenRouter Very surprised to see Llama models in this which are basically garbage. And not enough of Gemini and Gemma
2
0
3
It’s now 27 years since I was made redundant and given a few grand. I decided that if I was careful, I could live off the money for 6 months trying to become a comedian before I had to get another job. I was 37. Worth a punt.
4K
6K
137K
Can someone tag Emergent folks here? I can't use my US credit card and the Indian card does not work. I want to give them money and they won't accept it.
3
0
8
new writing: @ArizePhoenix has done a great job with Open Inference and I refuse to let it be misunderstood
0
1
7
RFC: more dspy modules or some other way to directly do agent optimization maybe something like variations across LLMs which lm itself as a parameter while I am at the wishing well, can someone please add docs to make dspy work with @ArizePhoenix
1
1
4
im so tired of the 30-page ai-slop brain-rot around RAG here - i solved it for you. this simple graphic tells you everything you need to know.
I wrote 6 months ago that RAG might be dead. That was after an aha moment with Gemini’s 1M context window - running 200+ page docs through it and being impressed with the accuracy. But I didn’t have skin in the game. @nicbstme does. He just published an excellent piece: "The
36
86
945
prefilling user feedback for optimizers is such a clever UX idea, I am jealous that I did not come up with this
DSPy / GEPA Pro Tip: Have your module signature output both a reason and a red-flags text field. In your UI add a thumbs up/down that users can click to agree or disagree with the result. If they disagree toggle the value. Based on the updated value, display a feedback field
0
0
5
From Dermatology NP to X/@JoinAutopilot_ Investor: How I Turned $100,000 into a Multimillion-Dollar Home “I cannot live without books.” - Thomas Jefferson I didn’t grow up in a wealthy or financially savvy family, but I’ve always been a voracious reader. From a very young age,
4
1
30
hot take: the main barrier to late interaction adoption is the lack of widely trusted "vector store" equivalent elastic happened because lucene and solr worked. weaviate, pinecone and qdrant made hnsw usable. someone needs to make PLAID ready for atleast early adopters
0
0
2
Someone will point out that Super30 does not. I disagree: it exploits opposite extreme of the resources vs intelligence curve Super30, by definition, selects some of the smartest people in the under-resourced curve: Early intelligence, believe it or not — ovarian lottery
0
0
1
Stanford and Harvard Business School have legacy admissions. Indian Govt Jobs do this for proper, educated adults. We do this for elected representatives! The ovarian lottery (google this!) is the greatest lottery of all — and all institutions want to exploit it at some level.
Most schools still ask for parents’ job & education on admission forms. How does that decide if a child deserves an admission? Kids aren’t applying for a loan, they just want to learn. We’ve to get rid of these outdated biases. Not helping anyone.
1
0
1
DSPy / GEPA Pro Tip: Have your module signature output both a reason and a red-flags text field. In your UI add a thumbs up/down that users can click to agree or disagree with the result. If they disagree toggle the value. Based on the updated value, display a feedback field
14
23
374
Take your gaming experience to the next level with the cosmic Space Whale design.🐳 The K75 mechanical keyboard delivers smooth keystrokes with its custom mount and gold-plated stabilizers, while the M2 mouse ensures you're always on top of your game with lightning-fast
3
8
13
the best part about startups moving to zoho will be that in ~2 years they'll start cribbing how much it sucks (css, speed, uptime) and hence help make zoho better in 5-10y
2
1
23
There are only two skillsets which matter in the post AI world: 1. Making intros 2 Knowing how to make *great* intros
I am very, very good at intros. It’s a part of the job, but also something I just love doing. Nothing gives me more of a thrill than getting two people I admire in a room together. So here is my Master Intro Process: 1. Get that double opt-in. For some people (if you’re close
0
0
5