The third edition of AI Weekends took a different approach. Instead of another keynote, the room played bingo. By the time the session wrapped, 41 attendees had mapped 20 of the most important technologies in the modern AI stack — and actually understood how they connect.
A Different Kind of Session
Hosted at Coimbatore Institute of Technology's M114 block, Vol. 3 was designed around a simple observation: AI terminology has become a wall. RAG, LoRA, vector stores, guardrails, agents — the words are everywhere, but the mental model connecting them rarely is. The session set out to fix that.
Siddarth Kengadharan, founder of Facilitron, led the room through what the team called AI Bingo — a framework that organises core AI technologies into five categories: Brains, Memory, Actions, Blueprints, and Safety. Attendees received scorecards at the door.
How AI Bingo Worked
The format was deliberately low-stakes and high-engagement. Siddarth introduced each technology through a real-world application — not a definition, but a story. A customer support bot that remembers context across sessions. A coding agent that searches the web before answering. A production model that refuses certain outputs without explanation.
As each example landed, attendees identified which cell on their scorecard it matched and marked it off. The competitive element kept the room's attention sharp — but the real effect was that every technology got anchored to a concrete use case before any jargon was introduced.
The five categories did real conceptual work. "Brains" covered foundation models and inference. "Memory" covered vector databases, retrieval, and context management. "Actions" covered tool use, APIs, and agent capabilities. "Blueprints" covered orchestration frameworks and prompt engineering patterns. "Safety" covered guardrails, evaluation, and alignment techniques.
Two Sprints, One Picture
The session ran across two one-hour sprints with a break in between. The first sprint covered the foundational layer — models, memory, and the basic plumbing of an AI system. The second went higher up the stack: how orchestration frameworks tie it together, how safety mechanisms sit at every layer, and what "production-ready" actually means for an AI application.
By the end of the second sprint, the scorecard wasn't just a game prop. It had become a rough architecture diagram — a map of how the technologies relate to each other, built up incrementally through stories rather than slides.
Who Was in the Room
Vol. 3 drew a deliberately mixed crowd: developers who knew the code but not the landscape, product managers trying to scope AI features intelligently, and founders evaluating where to place bets. The format served all three. By the end, the developers had vocabulary for conversations with stakeholders, the PMs had a framework for prioritising, and the founders had a clearer picture of the build-vs-buy decisions ahead of them.
The session closed with an open floor where Siddarth fielded questions ranging from the practical ("which vector database should we start with?") to the strategic ("when does fine-tuning actually make sense?"). The answers were honest about trade-offs — which is exactly what the room needed.