From 10,562 teams to Top 65. Here's what we learned building an AI assistant in 24 hours.
At 2 AM, testing our real-time document analysis, my teammate asked: "What if this actually scales?"
We were building a Due Diligence Copilot, an AI assistant using Pathway that helps investors analyze financial documents in real-time. Not just another hackathon project, but a solution to a real pain point in M&A deals.
The challenge: CodeSlayer 2k25 by DevSphere India at NIT Delhi. Our track: Due Diligence Copilot (Pathway AI). The reality: 10,562 teams competing, only 65 would make it to finals.
What we created: an AI assistant for real-time document analysis using Pathway, high-accuracy citations from financial statements, a live dispute resolution system, a dynamic knowledge base that updates as new files arrive, and a production-ready architecture for investor teams.
The journey involved sleepless nights optimizing ML models, countless iterations on our demo flow, team debates that made our solution stronger, and that moment when our live file integration worked perfectly.
The result: we made it to the Top 65 finalists (0.6% selection rate). We didn't take home the trophy, but we gained proof that our technical skills can compete nationally, experience building AI systems under intense pressure, a network of incredible innovators, and the confidence to tackle complex problems.
Key learning: when you build with technical excellence and purpose, you never truly lose.
