Umang Lal

Tweets are personal views | Retweets are not endorsements | Learner ✌

[Umang Lal @Umang_lal]

Not sure if she is sarcastic or serious with localhost..

[Sarah @araseb_]

Claude Fable 5.1 is insane.

i know literally NOTHING about coding. ZERO.

and i just built 3 fully functioning web apps in 30 minutes.

http://localhost:3000/
http://localhost:8000/
http://localhost:5000/

check it out.

[Umang Lal @Umang_lal]

@Paytm code push without QA?

[prathosh A P @prathoshap]

With the Blessing of Sri Krishna, I am launching Bhāgavata-VāNi (Link next post)

The complete Śrīmad Bhāgavatam (all 12 skandhas) with synced audio recitation and karaoke-style highlighting.

- 10 scripts with topic descriptions
- ad-free
- stotra list,
- advanced searches

[Pratyush Kumar @pratykumar]

Ok, so here is my take on the Fable ban, sovereign AI, Sarvam, etc.

The event is interesting as it has implications from many perspectives.

For AI users, it is clear that you should not confuse access with ownership, or adoption itself as advantage. And if the most significant tech differentiator you are leveraging has external control loops, then you have to accept you are vulnerable.

For AI talent, it is now a precedent that you would be *seen* aligning to national interests more than company interests. And even if its just a whim for now, this trend will be hard to reverse as the world gets more automated…

For AI labs, their offerings will be stratified - general purpose AI would be available as utility, but frontier AI would be gated. This is a fantastic business model for labs - *democratized* AI sucks in all the data liquidity of the world which is locked in higher margin frontier offerings.

I think for the world to be a better place, all three of the above are bad vectors. We need to have more countries and companies owning their own destinies. And in the post AI world, that means being able to use and improve AI systems within their own perimeters - what one may call Sovereign AI.

At Sarvam, Sovereign AI in India was the founding thesis a couple of years back, and continues to remain the core operating principle. From our vantage point, it is super clear that India will build, leverage, and create massive business value and societal impact with sovereign AI. The following is precisely how we at Sarvam are contributing to make that happen.

[Umang Lal @Umang_lal]

@anshb24x7 ++

[Umang Lal @Umang_lal]

Krishnavataram ❤️
Should be publicized more! #krishnavataram

[Umang Lal @Umang_lal]

Had a good time building using @anakinHQ at the mini hackathon :)

[Sarvam @SarvamAI]

We are excited to announce that Sarvam is partnering with @PixxelSpace to power the AI backbone of India's first orbital data centre satellite.

This is a first for the country, with India-built AI models running on an India-built satellite and both training and inference happening directly in orbit, without any dependence on foreign cloud or ground infrastructure.

[Beats in Brief 🗞️ @beatsinbrief]

🚨 BREAKING:

Nvidia, Accel & HCLTech in talks to invest $200–250M in Sarvam AI at $1.5B valuation, potentially making it India’s first AI unicorn of 2026. (ET)

[Pratyush Kumar @pratykumar]

Delighted to be a founding member of @nvidia's Nemotron coalition. Bullish on what the coalition can achieve together. And India's sensibilities of being voice-first, linguistically diverse, and cost efficient at population-scale will be at the global table shaping frontier research!

[Sarvam @SarvamAI]

Listen at Scale is a joint initiative by Sarvam, @EkStep_Org, and @ai4bharat to deploy our multilingual Voice AI agents at population scale across India.

Hundreds of millions of people cannot navigate apps or text-based systems, and traditional outreach only ever broadcast at them without listening back. This initiative was built to change that.

In just 31 days, working with 20 partner organisations in healthcare, agriculture, governance, and skilling, our multilingual Voice AI agents reached 50 lakh Indians and logged 74 lakh minutes of real two-way conversation.

Read the full report: https://t.co/ut3n6xSkz3

[Umang Lal @Umang_lal]

@anshb24x7 @Strava Pretty sure about that 🔥

[Elon Musk @elonmusk]

People giving OpenClaw root access to their entire life

[Aakash Gupta @aakashgupta]

Sarvam went from “embarrassing” to “well done” in eight months, and the reversal is coming from the same VC who trashed them.

Deedy called Sarvam’s flagship LLM launch embarrassing in May 2025. Said their 24B model got 23 downloads while two Korean college kids got 200K. Now he’s saying they have the best TTS, STT, and OCR models for Indic languages. Same person, same company, completely different verdict.

The part nobody’s talking about is what actually earned it.

In May 2025, Sarvam was trying to play the LLM race. They fine-tuned Mistral Small, called it Sarvam-M, launched it on Hugging Face, and watched it flatline at 334 downloads while the Indian tech community tore them apart. $53M in funding, government backing, 4,096 H100s from the IndiaAI Mission, and their flagship product landed with a thud.

Then they stopped chasing parameter counts and started building the applied stack: Bulbul for text-to-speech, Saaras for speech-to-text, Sarvam Vision for OCR and document intelligence across 22 Indian languages. Products that solve actual problems in a market of 1.4 billion people, 800+ million of whom don’t speak English.

Google, OpenAI, and Anthropic allocate 1-3% of their training data to Indian languages. They think in English and translate out. Sarvam is building natively for Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, and 16 others. Direct Indic-to-Indic translation without routing through English as an intermediary. A fundamentally different architecture choice that big labs have zero incentive to replicate for a market they view as secondary.

Sarvam Vision launched two days ago and already outperforms Gemini 3 Pro on Indic OCR benchmarks with 87.36% word accuracy across 22 languages. Bulbul V3 dropped today with the highest listener preference in independent human studies.

Training the 500th LLM gets you 23 downloads. Building the best voice and vision stack for a billion people who can’t use Western AI products in their native language gets you a VC reversal in eight months.

Big labs will never go deep on Kannada OCR or Marathi speech synthesis. The ROI math doesn’t work for them. It works for Sarvam because they’re the only ones doing it well, and the Indian government is literally paying them to scale it.

[Umang Lal @Umang_lal]

Was expecting these great results from @SarvamAI . Huge!

[Pratyush Kumar @pratykumar]

Drop 2/14: Sarvam Audio: a state-space based efficient audio language model that defines the new benchmarks in speech recognition for Indian languages. Significantly outperforms Gemini 3 and GPT 4o Transcribe in a range of benchmarks. See details in our blog: https://t.co/GFcklhteKJ
@SarvamAI

[Pratyush Kumar @pratykumar]

Drop 2/14: Sarvam Audio: a state-space based efficient audio language model that defines the new benchmarks in speech recognition for Indian languages. Significantly outperforms Gemini 3 and GPT 4o Transcribe in a range of benchmarks. See details in our blog: https://t.co/GFcklhteKJ
@SarvamAI

[Umang Lal @Umang_lal]

@prerna__lohia Happy New Year !

[Umang Lal @Umang_lal]

@karthikponna19 Java

[Umang Lal @Umang_lal]

@prerna__lohia All the best! Hope you achieve everything you are wishing for :)

[Muhfaad @VastavikMuhfaad]

Music video releasing this 15.12.25 👑