My girlfriend introduced me to Amit Mishra, Founder & CEO of iMocha — an international HR-Tech company — and that introduction led to an invitation to their internal AI Hackathon. I signed up without hesitation.
I prepared the way I used to prepare for school competitions. The theme was AI, the tool of choice was Claude Code, and you don’t say no when someone hands you access to the Claude Sonnet model family. Got inspired by Steipete (founder of OpenClaw), set up Claude Code the night before, and showed up at iMocha by 8AM ready to build.
The hackathon ended at 10PM. Problem statements dropped at 9AM. We had 11 hours.
Team MANAV: built like it was meant to be.
Roshan Nangare · Mudit Dixit · Namita Khandare · Malhar
2 from Sales. 2 from Tech. A combination that turned out to be our biggest unfair advantage.
Sales visibility is reactive. Risk signals are buried in CRM notes, emails, and meeting transcripts. Action items and forecast accuracy depend entirely on human judgment — and humans miss things.
The challenge: build an AI Copilot that surfaces deal risks in real-time, calculates win probability dynamically, auto-generates next-best actions, clusters objections from call transcripts, and flags forecast inconsistencies before it’s too late.
I had never touched sales in my life. No process knowledge, no domain context — nothing. And here I was, building an AI agent for account managers.
Roshan walked Namita and me through the entire sales cycle: HubSpot deal stages, how clients engage with iMocha, how opportunities go cold sitting idle in email threads, where risk compounds at each stage. By 10:30AM we had domain clarity.
Then we did something most hackathon teams skip: we built the story before we built the solution.
If MANAV was already live, how would a salesperson’s day actually look different? We worked backwards from that. Three chapters. Three features. Three direct painkillers. We had the presentation narrative ready before a single line of code was written.
At 12PM we made a deliberate call: no shiny feature-rich BS that no one would actually use. We locked in the Minimum Evolvable Product. And here’s what blew my mind — Roshan, a sales guy, had already vibe-coded a proof-of-concept UI using Claude Code, telling us exactly what he needed as a customer. Product management is changing. The customer is now the first prototyper.
We split the product into two parts:
Integrations in scope: HubSpot, Microsoft Teams, Outlook, Fireflies.ai, and Telegram for demo.
Before touching any code, we dropped our entire context — screenshots, POC app, evaluation criteria, problem statement — into a research prompt on Perplexity. The output gave us a structured build plan and master prompts for both the Lovable dashboard and the Claude Code instance. Namita and I parallel-launched our builds from those prompts. Zero idle time.
One smart move: we used Claude to do API exploration on HubSpot and other live data sources. This gave us a near-real view of actual deal data — using which we generated a rich mock_deals.json and a data_dictionary.md. That became the backbone of our FastAPI server connecting the agent to the dashboard.
We ran two Claude Code instances simultaneously — one debugging the dashboard, one iterating on the agent. By 6PM we had a working mock demo. Mudit then wore the product hat and pointed out high-impact, low-effort bugs across the prototype. We iterated fast.
The evaluation criteria wasn’t “who built the best working prototype.” It was who used Claude Code and AI tools most thoughtfully. Are you thinking alongside the AI, or just jumping into solution mode? That distinction makes all the difference.
The energy at iMocha — the leaders, the participants, the entire environment — was something I hadn’t experienced at a hackathon before. Inclusive, ambitious, and genuinely curious.
A good AI sales copilot forces you to understand sales better than most salespeople do.
To automate something, you must first decompose it fully. By building MANAV, I ran discovery, pipeline design, objection modelling, and stakeholder empathy — in 11 hours. That is the engineer’s secret advantage in sales: you don’t just learn the what, you reverse-engineer the why.
Built with Claude Code · OpenClaw · Lovable · Anthropic API · HubSpot · Fireflies · Telegram
Team MANAV: Roshan Nangare, Namita Khandare, Mudit Dixit, Malhar