This isn't another "AI is magic" talk. It's a look under the hood. Every time ChatGPT finishes your sentence, it isn't thinking. It's doing math, very fast and at a huge scale.

Everyone uses LLMs every day. You prompt, it answers, and it feels like the model understands you. But here's what most people never see: behind every reply is a pipeline of vectors, dot products and probabilities picking one token at a time.

If you know that math, you stop guessing why a model behaves the way it does. You start to understand it.

Two kinds of AI users:

  • Prompt users: they type and hope. When the output is wrong, they have no idea why.

  • Builders: the people at this session, who learn how text becomes numbers, how context gets weighed, and how the next word actually gets chosen.

What we'll explore

  • Vector embeddings: how words and tokens are turned into numerical vectors, and why "king" and "queen" end up close together.

  • Cosine similarity: how the direction of vectors measures meaning, and how search, recommendations and retrieval (RAG) rely on it.

  • Attention mechanisms: how LLMs read context and work out which tokens matter to each other in a sentence.

  • Next-token prediction: how probability distributions and the softmax function help a model pick the next token, and what "temperature" really changes.

  • Behind the math: a simple, practical walkthrough linking linear algebra, vectors and probability to modern AI. No PhD needed.

The rundown

  • 05:00 PM | Samosa & Chai: networking with builders, developers and students.

  • 05:20 PM | Words to Numbers: tokens, embeddings and cosine similarity, explained visually.

  • 05:50 PM | Paying Attention: how attention turns a pile of vectors into context.

  • 06:20 PM | Guess the Next Word: softmax, probabilities and temperature, worked through live.

  • 06:45 PM | Open Floor: Q&A, discussion and connecting with the community.

  • 07:00 PM | Wrap-up

Who should comeAI enthusiasts, developers, students and curious learners: anyone who uses LLMs and wants to know what's really going on inside. If you remember a little school maths, you'll follow along.

The perks

  • 🧠 Clarity: leave with a mental model of LLMs you can actually explain to someone else.

  • 🤝 Community: Bangalore's AI builders, engineers, founders and students in one room.

  • ☕ Samosa and chai: because good math needs good fuel.

The people who get ahead in AI aren't the ones who prompt the most. They're the ones who understand what happens after they hit Enter.

Hosted by

OJone: a Pre-Series A startup foundry and incubator founded by successful entrepreneurs and technology industry leaders with global experience. OJone partners with early-stage startups in B2B, B2C, Deep Tech, Green Tech and Infrastructure and takes them through to the growth stage. Its builder pipeline runs from SamosaCh.AI, its curated, high-signal knowledge sessions and meetups, to Hackathons, Demo Day and AI House. Based in Bengaluru, with meetups in San Francisco, OJone is where people with knowledge become founders. Build. Break. Rebuild better.

AI House: the community layer powering this ecosystem. AI House is built for founders, researchers, engineers, students and AI enthusiasts. It's a space where people keep learning, sharing, collaborating and growing together. Through initiatives like 100 Days of AI House, community gatherings, founder conversations, knowledge-sharing sessions and industry collaborations, AI House has built a thriving network of 30,000+ community members who are actively shaping the future of AI. More than a community, AI House is where meaningful relationships, opportunities and ideas are created every day.