Sessions

The first session after the hackathon, spent entirely on what people had built there. Three projects, taken apart properly rather than praised. Ghormoi, from Garimaa's team, where the question was less about the code and more about whether the app made sense to the person opening it for the first time. Dhara, the inventory management system from Shivani's team, where we went from what it does today to what breaking it at scale would actually take. And Raghav's wedding website, which did not work on the day. So the room debugged it, then worked out what it would take to make it something people could actually use.

3 photos →

Sixty people came to The Bhawanipur House on a Sunday to build something in a day. Teams were assigned that morning, so nobody had to arrive with one, and most people met their teammates at the table. The rule was simple: by the end of the day you demo something that runs, not something you plan to build. The winning team built an edtech site for teachers, and two of its members are thirteen years old. They led it. Judges came from Indus Net Technologies, Navsoft and Navam Capital. The ₹50,000 prize pool was split ₹25,000, ₹15,000 and ₹10,000 across the top three. Jai Balaji Group and Cuepilot backed the day as co-sponsors, Indus Net put up the prizes, CDC Printers handled print, and Alchemyst, ROXX, Healthy Eats, Evrleaf, The Social Capitalist and The Bhawanipur House covered everything from credits to the room itself.

39 photos →

Scheduling and routines. Getting work to run on a clock instead of whenever you remember it, connecting email so Claude can send the thing itself, and putting your own recurring tasks on rails. We also played Wheel of Fortune with AI words, which got more competitive than anyone expected.

2 photos →

The first co-building session. Everyone built a chatbot on their own voice, their own quirks, the way they actually reply to people. The lesson underneath was context. We used supermemory to manage context over models.

2 photos →

Agent loops, and why "the agent says it's done" means nothing. We played Don't Trust Done: real-looking agent transcripts on screen, teams bet ship or loop on each one, points for calling it right. The rounds were built from real failure patterns, tests the agent graded itself, migrations that "ran without errors" nobody verified, completion signals fired early. The takeaway wrote itself. A stop condition the agent controls is not a stop condition.

1 photos →

We built the thing Shivani actually needed. An inventory management system, with a CRM wrapped around it so there was a real use case to point at rather than a demo. Built in Lovable, then picked apart as a room, what was missing, what would break, what to do next. We also played Jeopardy.

4 photos →

First we started with Lovable. Vidisha has a design agency but not a website, so we used Lovable to build a website for her. She described her exact vision and the output came pretty close to whats she wanted. Also Dipika, who is an influencer, used Claude to make a mini-app for tracking brand deals.

1 photos →

Image and video generation, hands on. The main event was the Image Auction: ten AI-generated images of Kolkata went under the hammer, every team started with ₹1,000, and you bid before seeing the prompt. Only after the first hammer fell did the room learn the score: total word count of the prompts you won. At least one team paid heavily for a three-word image, which was the whole lesson. Telling a three-word prompt from a paragraph one, by eye, is a skill, and now everyone in the room has priced it with fake money.

3 photos →

We discussed what skills everyone was using, and we even built a writing skill on spot - that helps you write in a way that does not sound AI-generated.

1 photos →

Our first hackathon. Three hours to build, one hard rule: post a live URL in the WhatsApp group by 2:45. 6 teams showed working products to the room. Judging was blunt: does it work, is the problem real, would someone use it this week. Code quality was never a category. Team Homie's product took the prize, and one build outlived the day entirely: the member directory that became calaiclub.com, cream and ink and all, started as a hackathon project in this room.

3 photos →

Lot of new faces showed up for this one, but the talent still remained top-notch. Again, some were new to AI and some building heavily for a while. We met Krishnav, who is deeply knowledgeable about vibecoding and has built a few great products. Ahan used Claude to help with 3D modeling of products.

3 photos →

Around 12 people showed up. Some building with AI hands-on and some wanting to learn about it. Ayush Jhunjhunwala and Yash Rathi gave great demos and taught the others about interesting tools.

6 photos →

Around 10-12 people showed up, all of them building in AI. Some were building internal systems for their family businesses, some providing services and operating as agencies, and one person even got funding for his startup.

3 photos →