HOW I BUILT A $4M BUSINESS
Without knowing farming. Without knowing how to code.
THE GUILT
June 2020
I was in my dorm room watching Game of Thrones. Outside, thousands of farmers were marching to the state capital. Drought. Crop failures. Suicides.
I kept watching the show.
The next day, the guilt wouldn't leave. I'm an engineer. I should be able to help. I submitted a project proposal. Zero knowledge about agriculture. Zero connections. Just guilt and a vague idea.
> proposal.submit() STATUS: APPROVED knowledge_farming = 0 knowledge_coding = "basics" connections = [] // They approved it. // Now I had to actually do something.
EVERYTHING FAILED
Build an App
Farmers can use it to get better prices.
ERROR: The ones who do don't trust random engineering students.
Sell Pesticides
Better pesticides = more profit for farmers.
ERROR: Local agents made money on volume, not outcomes.
THE MUSKMELON MOMENT
Summer 2021
I saw a farmer on TV. He grew muskmelons. Sold them for 10x what other farmers got for regular crops.
"Premium crops need premium buyers. I found a buyer who cares about quality. Now quality is profitable."
That's when I understood.
The problem wasn't that farmers didn't have technology. The entire incentive structure made quality unprofitable. Middlemen bought by volume. Retailers wanted cheap. Farmers had no reason to grow better crops.
- Middlemen buy by volume
- Retailers want cheap
- No incentive for quality
- Farmers get premium prices
- Buyers get consistent quality
- I get coordination margin
FIRST DEAL: 5 TONS OF MUSKMELONS → LOCAL RETAILER
STATUS: IT WORKED
SEEDLESS WATERMELON MONOPOLY
The muskmelon model proved: premium crops + premium buyers + direct relationships = profitable for everyone.
Seedless watermelons. India didn't have them at scale. Every grocery store imported them or didn't stock them. Farmers who tried to grow them failed because they didn't know the technique.
I spent 3 months learning seedless watermelon cultivation. Found farmers willing to experiment. Taught them the technique. Built relationships with Reliance, Namdari, and premium retailers.
{
"year": "late_2021",
"achievement": "India's first seedless watermelon supply at scale",
"farmer_earnings": "3x regular watermelons",
"retailer_satisfaction": "consistent quality",
"competition": "none",
"reason": "no one else figured out the full system"
}THE DO-OR-DIE MOMENT
September 2021
Phone rings. Lulu Hypermarket Dubai. They want 50 tons of watermelons. Massive order. Could change everything.
I said yes.
Then Maharashtra started flooding. My farmers couldn't harvest. My usual contacts couldn't deliver. I had never been to Karnataka. Didn't know anyone there. Didn't speak Kannada.
I went to Karnataka.
Cold called farmers. Showed up at farms. Explained what I needed. Built trust in days, not months. Found farmers in Andhra Pradesh. Then Telangana. Then Tamil Nadu.
IN 2 MONTHS: BUILT 4 SUPPLY CHAINS ACROSS 4 STATES
DELIVERED 50 TONS TO LULU. ON TIME. CONSISTENT QUALITY.
They kept ordering. Then Reliance. Then Namdari. Then Zepto. Then Swiggy.
THE OUTCOME
The business worked because I understood something most people missed:
WHAT THIS TAUGHT ME ABOUT AI
AI is in the same place agriculture was in 2020. Powerful tools exist. Most implementations fail. Not because the technology doesn't work. Because the incentive structures are broken.
Builders optimize for demos, users need reliability
Users burned by AI hype, hesitant to adopt
AI tools that don't fit into actual workflows
This is why I'm learning AI from first principles. Not to build demos. To build systems that actually work for the people using them.
EPILOGUE
I graduated in 2024. Came to NYU for my MS in Computer Engineering. Now I'm learning how frontier AI models actually work. Transformers. RLHF. Chain-of-thought reasoning. RAG systems. Interpretability.
From first principles.
Same velocity I learned farming. Same method I built supply chains.
SUMMER 2026: LOOKING FOR AI ROLES WHERE I CAN BUILD PRODUCTS THAT SHIP TO REAL USERS, WORK IN MESSY 0→1 PROBLEM SPACES, AND MOVE FAST.