Shubham Kumar

Building : Never miss a call
Agents and human on the same call !!

Looking for Intros to people, operators and angels.

[Shubham Kumar @Shubham04jan]

@phimaanshu Hi, would love to try

[Shubham Kumar @Shubham04jan]

@ayaboch Also find Emergent

[Deedy @deedydas]

Meritocracy in tech is precarious. It’s everywhere and nowhere at the same time.

Cynics will tell you there is none. How can it possibly be fair that the Stanford grad seemingly can raise from any VC they want but the VirginiaTech one has to struggle? If you’re from a big lab, you’re the prom queen, but if you’re from a yesteryear SaasCo, no one even picks up your resume. Warm intros through people in the know seem to dictate who is allowed to win. To even get an interview at YC or a VC gig, it feels like an elite education is a prerequisite. It can seem like the system is rigged against you.

Yet, that view would be so patently false. The valley is one of the best meritocracies in the world. It’s just not obvious when you’re young. Tim Cook went to Auburn and spent 12yrs at IBM before Compaq. Nikesh Arora, one of the highest paid execs in America, went to a “lower ranked” IIT from India. Jan Koum dropped out of San Jose State to work at Yahoo before founding Whatsapp. The CTO of Anthropic went from a no-name college most couldn’t put on a map. Dylan Patel who sits on the most influential set of data in the world, got there by pure passion. Dwarkesh was a nobody before running the most influential podcast in tech. The team that built GPT-3 isn’t exactly a Harvard reunion. Jensen was from Oregon State and Denny’s. Tobi Lutke, Palmer Luckey, Frank Slootman, John Carmack, Jeff Dean, Linus Torvalds, Satya Nadella, the examples are infinite.

How can both of these be true? At any given moment, it feels like a stratified society based on your current credentials. What the Valley offers is the fastest path to updating those credentials.

The unspoken rule of the valley is that you get infinite shots on goal, but you just have to prove yourself to buy the next shot. If you try to jump too high, you might hit a brick wall. For some, it’s easy. You went to Stanford / MIT. Everyone thinks you’re the sharpest of your generation. But opportunity is abundant and you should be risk-on. One amazing product, one viral open source repo, one phenomenal company you helped build, one highly technical blog post, one amazing career stint: all can shape your fortunes overnight.

We often forget how much of a privilege that is. In most other parts of the world, you cannot in fact work yourself into the elite strata. CEOs of companies in India or China rarely are from a non-elite university. Corporate structures there are far less accommodating to former failures. Even in New York finance, there’s a certain way you need to speak, attire you need to wear and an unspoken private club that needs to welcome you. That doesn’t exist in the Valley. Sure there are private clubs, but anyone can make it to them. But you can always buy your entrance ticket with proof of work.

In the moment, it can seem not meritocratic. This happens especially when you’re younger. Hell, it’s not perfect and luck may never strike you. But career is a long game, and a wise valley veteran once told me “if you’re ambitious and you spend 25yrs here, there’s next to no chance you’re not making it.” And I’ve seen that too. It might not be fair in the moment, and it may not be fair to everyone, but it converges very quickly to it.

[Shubham Kumar @Shubham04jan]

@manavj29 Rama books are like talking to a jarvis about India !

[Shubham Kumar @Shubham04jan]

@xuster Compute bills !!!

[Shubham Kumar @Shubham04jan]

@brijbhasin Building https://t.co/scwzV43O0X would love to collect your experience working with voice agents.

try our TTS https://t.co/uTW2iMrQjJ

[Shubham Kumar @Shubham04jan]

@kushgrwl Infra and model both are required.

If we have hyperscalers and no frontier model then aws, azure, gcp will the customers for there data centers which will deploy frontier model and sell back to us !!!

[Shubham Kumar @Shubham04jan]

@rajeshsawhney In longer run Flipkart and Amazon will lead as they have cashcow ecommerce to fuel qcom.

[Shubham Kumar @Shubham04jan]

@rajeshsawhney Minutes is building on large scale ! Having Worked for both amazon and flipkart, Flipkart is aggresively building vertically Qcom in FK fashion, MNow (200+ darkstores) , Minutes (1200+) with existing users base and penetrating in T2 cities.

[Shubham Kumar @Shubham04jan]

The one rule you can't skip: judge ≠ target model family.
LLMs favor their own generations (the "self-enhancement bias" cited in the Nubank paper).

If the target runs GPT and the judge is GPT, the judge tilts toward passing it. Different family, or an ensemble/cross-model panel which is exactly what the Nubank cross-model κ analysis is checking for. Same logic applies weaker to the caller: if caller and target are the same model they share priors and fall into an unrealistically smooth groove.

Different model = more friction = more realistic.

reason all models are required #chatgpt #claude #deepseek

[Shubham Kumar @Shubham04jan]

When you try to paste response from one LLM to another.

[Shubham Kumar @Shubham04jan]

LLMs optimize toward success. Trained to be helpful

What if we make neutral !!

@sama @elonmusk

[Shubham Kumar @Shubham04jan]

@coryalthoff Majority not all !

[Noam Brown @polynoamial]

The cost of generating the proofs for all 10 of these breakthroughs combined was under $2,000 at Sol API prices. We’re excited to see what scientists and researchers are able to create with our upcoming Astra models!

[Shubham Kumar @Shubham04jan]

1/ Every voice AI company has a gross margin problem it isn't talking about.

2/ SaaS was the business of copying a string. Write it once, the second copy is free, 85% margins are normal. Per-seat pricing, land-and-expand, spending a year's revenue on CAC — all of it sits on that one fact.

3/ Voice doesn't copy. It computes. You can't write the prompt once and sell copies of the output. Every minute of every call costs real money to serve, forever. The marginal cost isn't near zero — it's most of the price.

4/ So the whole category is heading the same way: lower gross margins, thinner net margins, much more scale. Less Salesforce, more Walmart.

5/ In that world there are two seats. You rent your cost base or you own it. Almost everyone rents — their COGS is ElevenLabs' revenue, and it improves when ElevenLabs decides it does.

6/ In India it's not even a strategy question. A call has to land near ₹3–4/min before a 20-person business will buy. You cannot get there paying ₹2–6/min for speech alone. Owning the model isn't an edge here, it's the entry ticket.

7/ So we built our own. Telenow TTS: ₹0.40/min.

8/ Here's the part worth sitting with. An 85% margin business invites attack — anyone can undercut you and still make money. When you own the cost base and your competitor rents it, you can price below their cost. No funding round fixes that, because the money leaves every month as COGS.

9/ Hear it: https://t.co/yYfUUU5fBQ

[Shubham Kumar @Shubham04jan]

@SinghDevHub Try https://t.co/cdsDmNfd14

[Shubham Kumar @Shubham04jan]

@anuranjan @sridharfyi In case you are looking for voice agents do reach me for 10k free credits on https://t.co/scwzV43O0X

[Shubham Kumar @Shubham04jan]

@sidharthsajith5 @sridharfyi Do you use voice agents

[Shubham Kumar @Shubham04jan]

@archiexzzz Try https://t.co/cdsDmNfd14

[Shubham Kumar @Shubham04jan]

@sameerpallav @akshaymarch7 Try https://t.co/uTW2iMrQjJ