NirantK Profile Banner
Nirant Profile
Nirant

@NirantK

Followers
15K
Following
45K
Media
1K
Statuses
10K

AI engineer that takes companies from RAGs to riches

localhost
Joined December 2012
Don't wanna be here? Send us removal request.
@NirantK
Nirant
1 day
do we know how pepy gets pypi download data?.
1
0
0
@NirantK
Nirant
2 days
2 entries in 1 quarter to my personal Brag Doc β€” highest rate since 2021!. Might have done something right last year?. All hail my bae, my personal serotonin dealer with an assist from Duloxetine.
0
0
12
@NirantK
Nirant
2 days
The TER for this is ~1% as per @capitalmind_in website and 0% as per @CoinByZerodha
Tweet media one
1
0
1
@NirantK
Nirant
3 days
Hear me out: VC-funded VSCode-fork with first class native-support for LLMs which competes globally.
@NirantK
Nirant
3 days
Hear me out: VC-funded Chromium-fork with first class native-support for LLMs which competes globally.
1
0
6
@NirantK
Nirant
3 days
Hear me out: VC-funded Chromium-fork with first class native-support for LLMs which competes globally.
9
0
16
@NirantK
Nirant
4 days
Glad to see that @spacy_io Sentencizer remains a goated chunking approach β€” someone just used it to make 3B chunks!. It is also a phenomenal tokenizer fwiw!
Tweet media one
3
1
20
@NirantK
Nirant
5 days
Looks like Kazinama has been missing for a while now. No new music for a prolific account. I miss it
Tweet media one
0
0
2
@NirantK
Nirant
7 days
Medical and age-linked premium hikes are a double whammy β€” the price of services and goods increases, and so does the premium. So you lose when paying the premium and then again when you're claiming it πŸ˜….
0
0
3
@NirantK
Nirant
7 days
If you're paying 30% of the insurance amount as premium, at what point will the insurance be worth less than putting a 1.5x premium in a Liquid fund?.
@AdityaD_Shah
Aditya Shah
7 days
While people keep bragging about low inflation,. I just paid my parents' Medical Insurance premium. 3 years ago I paid,.1.72 lakhs for a 3-year medical cover for a 10L policy. This time the company asked for,.3L for 3 years medical cover for a 10L policy. That inflation is a
Tweet media one
5
0
6
@NirantK
Nirant
8 days
Spikes are all that matter after a certain kind of minimum.
Tweet media one
@surbhiskjain
Surbhi Jain
8 days
Average builds you quietly, spikes open the doors. Back in 2015, I spent days manually going through 1500 IIT Bombay CVs to shortlist a few for the company I was part of. Each was evaluated across ~5 sections, 3 points each. - Average work? 1 point. - Good work? 3 points. - An.
0
1
13
@NirantK
Nirant
9 days
Nailed it. Same risks as social media and WhatsApp in hindsight: Unknown, new stupid behaviours.
@paraschopra
Paras Chopra
9 days
I’m starting to believe that (current) AI systems pose risk to humanity not because they are super intelligent, but rather because they are super stupid in ways we can’t know upfront.
0
0
26
@NirantK
Nirant
10 days
If @openai was an Indian company .
Tweet media one
3
1
82
@NirantK
Nirant
12 days
So many Indian (and otherwise Made in Asia) apps use "Clone" instead of "Duplicate" like @zoho β€” cute!
Tweet media one
3
0
3
@NirantK
Nirant
12 days
@ajeetdsouza @vortex_ape @tokenbender Frappe (ERPNext?) by @rushabh_mehta and friends is Commercial OSS ERP, not just out of India but out Mumbai. Quite rare. I don't know their revenue as a business, and they do maintain the package: 43 contributors (incl. employees) in just last 1 month!.
@KapilSingaria
Kapil Singaria 🐊
12 days
@NirantK The Frappe web framework @ are by Indian devs. It powers (arguably) the most powerful open source ERP software. And many more.
0
0
3
@NirantK
Nirant
12 days
Enjoyed what I wrote? . I usually shit post here, so follow @NirantK if that's okay with you!.
0
0
5
@NirantK
Nirant
12 days
@ajeetdsouza @vortex_ape And the latest β€” one of the moderators of the GenerativeAI Community and a maverick in his own league @tokenbender: RL for smol LM experiments. It's clever, and pushes reward design in a new direction which atleast I've not seen outside academia!.
@tokenbender
tokenbender
12 days
is it possible to pretrain a language model using pure reinforcement learning from scratch?.random weights, no cross-entropy loss pretraining. you may have many questions in your head.
Tweet media one
2
0
8
@NirantK
Nirant
12 days
@ajeetdsouza Happy to include a former co-conspirator in @vortex_ape! .
@AkashTandon
Analytical Monk
12 days
@NirantK it's been a minute but shoutout to @vortex_ape who built Camelot to tackle gnarly pdf parsing back when he was still in the country and the problem was much more of a pain.
2
0
3
@NirantK
Nirant
12 days
@ajeetdsouza Kitty and Calibre come from the same dev in Mumbai as far as I know .
github.com
Cross-platform, fast, feature-rich, GPU based terminal - kovidgoyal/kitty
@DSantra92
Deeptendu
12 days
@NirantK For me, Kovid Goyal, the developer of kitty. Kitty is still my go to choice for a terminal emulator on any OS.
1
0
4
@NirantK
Nirant
12 days
zoxide by @ajeetdsouza has been maintained for almost ~4y now! pretty cool! .
@iavins
v
12 days
give a follow to my fren @ajeetdsouza . wrote and maintains zoxide (in rust, btw), it has 25k+ stars and the first thing i install on my system.
1
0
5