How Madhav Krishna Built Vahan Into India's Largest Blue-Collar Recruiter By Digitizing Trust
The Columbia-trained AI researcher pivoted from English-teaching chatbots to placing 40,000 workers monthly - more than any platform in India. His secret: embracing the middleman, not eliminating him.
In the winter of 2014, Madhav Krishna handed a plate of dal chawal to a 10-year-old boy in tattered clothes outside his parents’ Delhi home. The child hugged him in return. That moment changed everything.
Something just snapped inside me. I was literally in tears. Because I realized how fortunate I was to be handing the food out and not receiving it.
The statistic that haunted him: if you have a roof over your head, internet access, a college degree, and can afford three meals daily, you’re in the top 7% globally. Madhav, a computer engineer with a Columbia AI degree who’d spent seven years at US startups like Jetsetter, had won what Warren Buffett calls the “Ovarian Lottery.” He decided to use technology to break it.
A decade later, his company Vahan places over 40,000 blue-collar workers into jobs every month with a team of just 150 people, making it India’s highest-volume recruiter. The journey there required unlearning everything Silicon Valley taught him.
Check out the video of the conversation here or read on for insights.
The Three Pivots That Led to Product-Market Fit
Madhav’s first attempt was Lakshmi, a voice-based English teacher that workers could call to practice conversation. The engagement was extraordinary. One user, Santoshi from Karnataka, spent six hours straight repeating the same 10 questions. But vocational training institutes wouldn’t pay for it. They were incentivized by placement numbers, not learning outcomes.
Next came corporate training. Uber paid ₹10 lakh to train 10,000 drivers through WhatsApp. This was 2017, before WhatsApp had an official business API. Vahan built automation scripts that would “open up a WhatsApp instance on an Android emulator, type in the message, click the button, and send it.” Revenue scaled to ₹50 lakh annually, but long-term contracts remained elusive.
The breakthrough came from Dunzo in 2018. The delivery startup asked if Vahan’s WhatsApp bot could filter candidate leads. A simple text-based qualification bot cut Dunzo’s hiring costs by 30%. Swiggy called. Then Zomato. Inbound inquiries flooded in.
Training ultimately is really nice to have. It’s a vitamin in a way, whereas the real painkiller for these companies was to hire people.
Why Half a Million Users Were Almost Worthless
By late 2018, Vahan’s referral program went viral.
We were getting over a lakh users every 10 minutes.
Servers crashed. At peak, they had half a million monthly active users.
And it was almost worthless.
The example Madhav cites: WhatsApp. Despite 80% of Indian smartphone users opening it daily, nobody paid the ₹50 annual fee. Meta’s average revenue per user in India is one-hundredth of what it earns in the US.
More critically, blue-collar workers don’t trust purely digital job searches. Cars are still bought in showrooms. Equity trades happen through brokers. E-commerce only took off after Flipkart introduced cash-on-delivery.
In a country like India, engagement doesn’t always translate into a business. The Valley playbook is, build engagement, then that will translate into revenue. That doesn’t work in India.
Madhav made a contrarian decision: don’t eliminate the middleman. Digitize him.
The Agency Model: Weaponizing Local Trust at Scale
India has hundreds of thousands of small recruitment agencies controlling 60-70% of blue-collar hiring. Vahan built a two-sided marketplace. Large employers like Zomato and Swiggy give Vahan high-volume hiring mandates. Vahan distributes these to its network of 2,000+ local agencies via a mobile app with CRM tools, AI screening, and real-time tracking.
The economics are outcome-based. Qualifying a lead pays ₹5-10. A successful placement with retention milestones pays ₹1,500-3,000 for gig roles and ₹5,000-6,000 for manufacturing. Vahan only earns when someone gets hired and stays. Nobody in India operates at this scale - TeamLease, a traditional staffing giant, does 15,000 monthly placements compared to Vahan’s 40,000.
The timing was perfect. India’s quick commerce war in 2024-2025 created a “rider crunch.” Platforms like Blinkit, Zepto, and Swiggy Instamart faced 30% vacancy rates and 35-40% monthly attrition. Average delivery earnings jumped from ₹21,000 in FY21 to ₹28,000 in FY25. Vahan currently captures 95% of its volume from delivery roles.
Voice AI: Coming Full Circle with Better Technology
In 2024, Vahan integrated voice AI - returning to the 2016 Lakshmi concept, except the technology finally worked. The stack uses Deepgram and OpenAI’s Whisper for speech recognition, GPT-4o for conversation intelligence, and ElevenLabs for natural text-to-speech in Hindi and regional languages.
The AI recruiter costs ₹2 per connected minute versus ₹3-4 for humans and tripled recruiter productivity. The proprietary advantage: hundreds of thousands of hours of call recordings between human recruiters and candidates, used to fine-tune models for Indian accents, dialects, and code-switching.
The system even predicts retention with 70% accuracy.
Based on the signals we capture on the call itself, we’re able to predict whether the person will take up the job and stay.
The model might be reading emotional cues from speech patterns, though it’s a black box.
Madhav’s design principle for the bot: “When you meet a stranger in India, you ask where they’re from. You find common ground. The bot needs to do that—establish credibility through social proof.”
The Path to Profitability and Regional Expansion
Having raised approximately $33 million (including Series B funding from Khosla Ventures in September 2024), Vahan is expanding horizontally into manufacturing and retail, and geographically into Southeast Asia and MENA. The company projects EBITDA profitability at 3X current scale, likely within the next year.
The recent acquisition of L.earn, an upskilling platform from the defunct Temasek-backed GoodWorker, adds a training layer. As India’s gig workforce grows from 7.7 million (2020-21) to a projected 23.5 million by 2029-30, regulatory tailwinds like the Code on Social Security, 2020 are forcing formalization. Platforms must now contribute 1-2% of turnover to a Social Security Fund. Vahan’s digital record-keeping positions it as essential compliance infrastructure.
We’ve done our books. Like Amazon and Flipkart, the first category they cracked was books. For us, that was delivery. Now we’re adding other categories.
Madhav didn’t try to change Indian behavior. He optimized for it. In a low-trust market, technology doesn’t replace relationships - it scales them. That’s how you build an operating system for opportunity.
Listen now!
Other ways to listen:
Your Feedback matters
As always, I’d love to hear your thoughts! Whether it’s about this episode or ideas you’ve been playing around with, shoot me an email at ad@thepodium.in. Your feedback keeps these conversations going, and I’m always up for chatting about your startup ideas too.
Until next time,
Your Host,
Akshay Datt
Brilliant case study on language-first design. The insight about training AI on hundreds of thousands of call hours to handle Indian accents and code-switching is what separatesactual deployment from MVP demos. I've seen similar patterns in voice projects where the propriatary data moat comes from dialectal nuance, not the foundational model itself. The fact that retention prediction works at 70% accuracy from speach patterns alone is wild.