Startup Stream

This weekend, I sat down to watch Karpathy’s 3-hour deep dive into how large language models like ChatGPT are trained — and it turned out to be absolutely worth every minute. What surprised me most? It’s not just for AI experts.

Karpathy explains everything in a simple, clear way that anyone can follow — even if you're new to machine learning.
If you've ever wondered how AI actually works, this is the one video you shouldn't miss. It's insightful, beginner-friendly, and a perfect weekend watch for anyone curious about the future of technology."

From tokenization and attention mechanisms to model training and inference tricks — this isn’t just a talk. It’s the ultimate roadmap to understanding large language models.

Here’s how AI like ChatGPT is actually trained — in three essential stages:

1. Pre-training:

The model starts by learning from massive amounts of internet text — books, articles, code, Reddit posts — to build a general sense of how language works.

2. Supervised Fine-Tuning:

Next, it’s taught to behave like a helpful assistant by studying high-quality, human-curated conversations. This gives it purpose and structure.

3. Reinforcement Learning (RLHF):

Finally, the model is rewarded for good answers through trial and error. It improves by figuring out which token sequences work best — sometimes even inventing clever reasoning strategies.

But it’s not perfect! Karpathy also dives into the “sharp edges”:

Hallucinations (making stuff up)
Trouble with basic tasks like counting or spelling
Dependence on prompting techniques to stay accurate

>> Tools, better prompts, and smarter training are all part of making LLMs more useful — and Karpathy’s breakdown shows exactly how it's happening.

Most jobsite delays don’t come from late materials or missing permits. They start with a spreadsheet no one updated. A timecard that never got filed. A crew member whose certification expired last week — and no one noticed.

Lumber, founded in San Jose, is quietly fixing one of construction’s most overlooked problems: the workflows that hold the workforce together.

It’s not another dashboard or reporting layer. It’s a field-first platform that helps contractors manage time tracking, onboarding, safety, and payroll in one place — built with the realities of jobsite operations in mind.

As someone who leads eChai’s San Francisco chapter, I’ve spent years learning from founders. But few have shaped how I think about product, people, and quiet execution like Shreesha Ramdas, Lumber’s co-founder and CEO. He’s been a mentor — and watching him build this company has been a masterclass in listening deeply and solving slowly.

Building With the Industry, Not Around It

Lumber was founded by Shreesha Ramdas, a longtime enterprise SaaS operator who previously co-founded Strikedeck (acquired by Medallia). In 2022, Ramdas began speaking with general contractors, payroll managers, and safety leads, and kept hearing the same thing: there were too many systems, and none of them were built for field teams.

The company started with mobile-first time tracking — a system simple enough for crews to use from a phone or tablet, but powerful enough to integrate directly into payroll and compliance processes. From there, the product expanded: onboarding, licensing verification, safety documentation, and wage calculations across complex labor rules.

AI That Handles the Details Contractors Don’t Have Time For

In March 2025, Lumber raised $15.5 million in Series A funding, led by Foundation Capital, with participation from Tishman Speyer, 8VC, Sure Ventures, Carbide Ventures, and Firsthand Ventures.

The funding is helping Lumber invest further in automation. Its AI systems now assist with:

Unlike generic AI platforms, Lumber’s models are trained on construction-specific workflows. The platform integrates with existing software but operates as a control center for crew operations — syncing hours, licenses, safety data, and payments across multiple teams and job sites.

Construction’s Missing Layer

While tools like Procore help manage project documents and progress, Lumber focuses on the people layer: who’s on-site, whether they’re certified, and whether the company is protected. These are the workflows that don’t get headlines — but when they break, everything else slows down.

Construction is a $1.6 trillion industry in the U.S., employing more than 8 million people. Yet for many companies, field operations are still managed with email, PDFs, or last-minute phone calls. Lumber offers a platform that adapts to how jobsites already operate — not one that asks them to change.

What's Ahead

Lumber’s roadmap includes deeper integrations with ERP systems, enhanced support for union compliance, and more AI-driven tools for managing jobsite risk. The company is also expanding across the U.S., working with mid-size contractors and regional firms that often don’t have access to enterprise-grade workforce tools.

The company was briefly featured on the NASDAQ tower in Times Square yesterday — a moment of visibility for a startup that’s mostly focused on building for the people behind the scenes.

https://www.linkedin.com/posts/lumberfi_lumbertakeover-nasdaq-industryleaders-activity-7323359839743524865-SCy8?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAAI-esYBPGLfFiDlILo1FUJJoIov1NM9P10

For Lumber, success isn’t about splashy tech demos. It’s about helping the 6 a.m. shift get started without delay.

Farms drying before droughts. Wildfires flickering to life before they spread. Transformers overheating quietly before they fail. These are not problems visible to the naked eye, but they leave thermal fingerprints.

Ahmedabad-based SatLeo Labs wants to detect those signals early, from orbit. The startup has raised $3.3 million in pre-seed funding to build a constellation of thermal-first satellites that turn rising heat into real-time early warnings.

The round was led by Merak Ventures, with participation from Huddle Ventures, GVFL, Java Capital, IIMA Ventures, and PointOne Capital.

SatLeo Labs was co-founded by Shravan Bhati (CEO), Dr. Ranendu Ghosh (CTO and former ISRO scientist), and Urmil Bakhai (CSO).

Why Thermal, Why Now

Most Earth Observation startups focus on optical or hyperspectral data. SatLeo is betting on thermal infrared, particularly mid-wave and long-wave bands that detect heat shifts invisible to traditional sensors.

According to SatLeo’s official site, their satellites will feature onboard edge computing. This allows them to process thermal data directly in orbit and deliver faster, more actionable insights on the ground. The goal is not just to collect images, but to enable faster decisions in critical sectors.

Where SatLeo Stands Today

SatLeo Labs is currently in the engineering and payload validation phase, focusing on developing its first thermal imaging satellite platform.

The company specializes in capturing mid- and long-wave infrared data from low Earth orbit to provide real-time thermal intelligence.

This data is intended to serve sectors such as agriculture, infrastructure, and disaster management. The team is also expanding its technical staff and preparing for early pilot deployments. Their approach includes processing thermal data directly on-orbit and delivering insights through a cloud-based analytics platform.

A Platform Approach to Risk

Huddle Ventures describes SatLeo as a full-stack company. Its vision includes developing a cloud analytics layer that transforms raw thermal data into sector-specific tools for agriculture, energy, logistics, and infrastructure monitoring.

While no customers have been publicly named, the product direction suggests applications for both enterprise and government use.

What’s Next

The company is currently focused on payload validation, regulatory preparation, and team expansion. Its roadmap includes deploying its first satellite and onboarding early pilot partners.

As global temperatures rise and physical risks multiply, SatLeo Labs is working to make heat a usable signal. The team is building satellites designed not just to see the planet, but to understand when and where it might break.

Gujarat’s Space Policy Tailwind

SatLeo’s growth is happening in parallel with India’s spacetech surge. More than 190 space startups have launched since 2020, following policy reforms and increased collaboration with ISRO.

In April 2025, the Government of Gujarat introduced its SpaceTech Policy 2025–2030. The policy outlines capital support, grants, and infrastructure for satellite manufacturing and downstream analytics.

Reactions on social media

Sheetal Bahl, Partner, Merak Ventures

https://www.linkedin.com/posts/sheetalbahl_satleo-labs-raises-usd-33-mn-round-led-by-activity-7323297532409741312-L5C2?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAAI-esYBPGLfFiDlILo1FUJJoIov1NM9P10

Vinod Shankar, Founding Partner, Java Capital

https://x.com/vinod_shankar/status/1917480231931371614

Sanil Sachar, Partner, Huddle Ventures

https://www.linkedin.com/posts/sanil-sachar-34a7978b_the-moment-we-met-shravan-dr-ranendu-ghosh-activity-7323612857378471936-JKpY?utm_source=social_share_send&utm_medium=member_desktop_web&rcm=ACoAAAI-esYBPGLfFiDlILo1FUJJoIov1NM9P10

..

SatLeo Labs is part of a growing spacetech ecosystem in Ahmedabad, which includes Peersight, co-founded by Gaurav Seth and Vinit Bansal, building a constellation of SAR and AIS satellites for maritime surveillance.

Another notable player is Optimized Electrotech, which designs electro-optic surveillance systems for defense and aerospace use cases. Together, these companies reflect the city’s growing presence in India’s deeptech and satellite innovation landscape.

Dishant was sitting across from a salon owner who was drowning in WhatsApp messages. Haircut requests, reschedule pings, payment screenshots - all jammed into one never-ending thread. He missed a few slots. A customer walked away. Revenue slipped through cracks the size of a thumb scroll. It wasn’t incompetence. It was inefficiency.

And that’s what got to him.

The world had moved on to AI agents and full-stack CRMs - but the average small business owner was still toggling between Excel sheets and WhatsApp. “What if scheduling didn’t need another app?” Dishant wondered. “What if it lived inside the one they already use?”

Dishant isn’t new to solving real-world inefficiencies. Before CalendarFix, he co-founded Gradeazy, an ed-tech tool for simplifying exams and grading. He’s spent the last few years building lean products for people who don’t have time to figure out “tools.” That salon moment wasn’t just an insight — it was years of pattern recognition snapping into place.

Not another tool — a fix

Here’s a quick walkthrough of how CalendarFix works:

https://youtu.be/C6OFpEHULSA

CalendarFix didn’t start as a moonshot. It started as a fix — a simple one. A way to help solo entrepreneurs, coaches, and small teams do the one thing that often breaks their flow: booking time. Dishant had seen it too often — time lost, leads dropped, payments forgotten. All because traditional scheduling tools felt too bulky, too alien.

So he built CalendarFix to work where businesses already worked — inside WhatsApp. No new logins. No clunky dashboards. Just a link you send. It books, collects payments, sends reminders, even nudges folks if they don’t show up. A quiet, background engine that turns chaos into calm.

The curse of being too simple

But selling simplicity is hard.

“People thought it was just another Calendly,” Dishant says. “But this wasn’t about links. It was about flow.” He had to unlearn the startup pitch game. No fancy decks. No buzzwords. Just real use cases: a yoga instructor in Goa, a therapist in Ghaziabad, a French indie brand running WhatsApp demos.

Then came ProductHunt. CalendarFix quietly launched and — to their surprise — ranked #6 product of the day. Signups trickled in from Switzerland, France, the U.S. And in two months, over 1,200 bookings flowed through their system. No PR push. Just word of mouth.

Flow over features

What makes CalendarFix sticky is not the tech — it’s the philosophy. “We’re not trying to build the next super app,” Dishant says. “We’re trying to disappear into the tools people already use.”

Their team is small — part product nerds, part design thinkers. They care less about dashboard depth and more about use-case clarity. How do you book slots if you're a coach running five free sessions a week? What happens if your client pays late? Every friction point is an opportunity to remove a step.

Changing habits, one slot at a time

Still, the biggest challenge remains: shifting habits. Users default to what they know — DMs, Google Forms, even manual spreadsheets. Convincing them to switch means showing them what they’re losing every day. “We had to learn how to tell better stories — not about features, but about time saved,” Dishant says.

And the stories are rolling in — a makeup artist who never misses a client, a workshop host who doubled attendance, a solopreneur who no longer dreads scheduling.

Betting on the invisible future

The long game? Building for where messaging becomes business infrastructure. “WhatsApp isn’t just for chatting anymore,” Dishant says. “It’s where people work. We’re just giving them the tools to do it better.”

CalendarFix isn’t flashy. It doesn’t shout. But it solves. And in a world full of noise, that might just be the most underrated superpower.

You can try it out at calendarfix.com - or, more fittingly, just DM them on WhatsApp.

Today, eChai Ventures is hosting its multi-city Startup Demo Day - happening simultaneously in 10 cities across India.

The format is straightforward: early-stage founders pitch their startups to a room full of fellow builders, get real-time feedback, and connect with the community. These are local, low-key, and regularly hosted meetups - not pitch competitions.

What’s Happening Today

Bengaluru
→ Draper Startup House, 6–8 pm
→ Also in Bengaluru: eChai AI Day at Urban Vault, HSR Layout, with
Swathi Moorthy (ET), Garvit Juniwal (Glean), Saurav Gopal (Capria), Dharmesh Ba (Business Hero). Host: Radhika Mohta

Hyderabad
→ CoKarma, Durgam Cheruvu, 6–8 pm

Mumbai
→ DevX, Andheri East (11 am – 1 pm)

Gurugram
→ Nasscom CoE, 5–7 pm

Pune
→ Ideas to Impacts, 4–6 pm

Jaipur
→ J Startup House, 6–8 pm

Surat / Vadodara / Udaipur / Rajkot
→ All hosted at DevX Coworking spaces in each of these cities, 6–8 pm

Also Happening Today

Singapore
→ AI x FinTech Social — curated gathering of early-stage founders working across Southeast Asia.

What Happened Earlier This Week

Ahmedabad
→ eChai Startup Demo Day at GVFL (Thursday evening)

Bay Area (Santa Clara)
→ AI Founders Circle —  meetup with leading AI builders in the Valley.

What’s Coming Up

Sunday, Bangalore
→ Startup Open House at Draper Startup House — informal morning meetup with early-stage founders.

Sunday, Kolkata
→ Partner-led meetup with local startup operators.


eChai hosts these meetups regularly — across India and globally — featuring Demo Days, AI Days, and founder socials.

To register for any specific event, visit eChai.Ventures.