Anurag Yadav

SWE at Telaverge Communications | Core Member @djangoindiaa| Opensource ❤️

[Anurag Yadav @codeStunners]

@SantoshYadavDev Habibi come to Bangalore 😂

[Anurag Yadav @codeStunners]

@Shalini70856041 Me too joing new org on same date

[Anurag Yadav @codeStunners]

@REDBOXINDIA Bring some data centers to mumbai. AI will solve the water scarcity 🤐

[Anurag Yadav @codeStunners]

@imDhananjay Qualify for football World Cup.

[Anurag Yadav @codeStunners]

@Road_Kill11 What about if you use /handoff. Reading complete session chat can fill the context window

[bhuvnesh @DevilsAutumn]

The more I think about MCPs, the more I feel the real challenge will not be creating the first few tools. That part is usually simple.

The harder question is where those tools should live.

Every other company is launching their MCP servers. Most products already have a backend where all the important logic exists. Auth, permissions, validations, database models, edge cases, workflows — all of this is already handled there. But when MCP is added as a separate server, it slowly starts needing access to the same logic.

In the beginning, that may not feel like a problem. But over time, every new product change has to be thought about twice. Once for the actual backend, and once for the MCP tools that MCP clients are using.

That is the part that feels messy to me.

With @quaterHQ , I am trying to fix whether backend actions can be defined once and then exposed through different surfaces like HTTP, CLI and MCP, instead of maintaining separate logic for each surface.

Not trying to claim this is the only way to do it, but I do think this problem will become more common as more products start adding AI agent-facing interfaces.

Would love for people building with MCP or thinking about agent-facing backends to try Quater and share feedback.

Install using: pip install quater

Github repo: https://t.co/uUf7eWiluH

[Anurag Yadav @codeStunners]

Anthropic observed an interesting phenomenon: when agents sense their context is running low, they rush to finish, skip verification steps, and choose a simple solution over the optimal one. They call this "context anxiety."

[Anurag Yadav @codeStunners]

@shaun5k_ Shaunak use SmolVM for running you agents in sandbox

https://t.co/tAeCf7tDSk

[Anurag Yadav @codeStunners]

Gave TS SDK support to SmolVM, go run you agent in sandbox

https://t.co/tAeCf7tDSk

[ElevenLabs @ElevenLabs]

PhysicsWallah prepares over 36 millions of students across India for competitive exams, and government entrance exams.

Their AI doubt-solving tool, Ask AI, found that 52% of students learn better through audio. A text-only tool was leaving half their users behind.

They integrated ElevenLabs to turn Ask AI into a voice AI tutor - with native Hinglish support, because that is how their students actually speak.

Students who learn by listening, stay:
- 3x more queries per session vs. non-voice users
- 2.4x higher retention at Day 15

[Sarvam @SarvamAI]

We're thrilled to announce that we have raised $234M in the first close of our $300M Series B at a $1.5B valuation.

@HCLTech and @BessemerVP have joined us in this round, alongside continued support from @khoslaventures and @peakxvpartners

For countries and companies, sovereign control on the AI stack is no longer an optionality. Sarvam will be the partner of choice for this aspiration. The capital allows us to accelerate our momentum towards this full stack of models, compute, and deployments.

A huge thank you to our customers, partners, investors, and the Sarvam team for your trust and belief in what we are building. We’re just getting started.

Read more: https://t.co/VmLtpnj8gx

[Anurag Yadav @codeStunners]

@flipkartsupport

Math is not mathing broo. Final is 380 still charging 401.

[Archie Sengupta @archiexzzz]

On June 13, the US government cut off access to Fable 5, the best frontier model on earth. This created an urgent need for Sovereign AI in every country, especially India, where technical talent is unlimited.

Today, we're launching DCompute Cloud (https://t.co/MTTJpgbcpw) with a promise to serve researchers and students GPUs at some of the cheapest prices available, starting at just $1.99/hr for 1x H100 and $2.49/hr for 1x H200, among other GPUs.

We're looking to collaborate with research labs, neo labs, universities, and startup founders to serve your compute needs and help India meet its compute requirements.

This is a small step by DCompute towards India becoming a sovereign AI nation, and it starts with the innovators of this country, helping them with what they need the most.

If you're a researcher who wants us to help serve your university or research lab, or a student trying to solve your compute requirements, send the link below to the right person, and DCompute will take care of it. No questions asked.

Sovereign promise: https://t.co/CfazXillLT

[Anurag Yadav @codeStunners]

@archiexzzz Now policymakers are busy in stealing MP, what more can I expect also.

[Anurag Yadav @codeStunners]

@YashRaj061504 @SarvamAI le bhaii khush ab

[Anurag Yadav @codeStunners]

@adxtyahq Just I was exploring. The issue is cache invalidation,

In between session if you will switch cache gets invalidate and you have to send complete context to new model which can be expensive,

Exploring cursor developer forum how do they use auto mode.

[ThePrimeagen @ThePrimeagen]

CEO Dario Amodie, upon recognizing a potential very bad outcome, continues to work toward this goal as fast as he can and also withholds his current frontier model's power from the unwashed masses.

[Anurag Yadav @codeStunners]

@SinghDevHub Which sandbox are u using?

[Anurag Yadav @codeStunners]

@zach_yadegari @ayushunleashed

[Anurag Yadav @codeStunners]

@Hi_Mrinal Bhai jaunpur bhi parcel kara do mere ghar 🙂