The voices on stage in the City of Lakes.

The talk

Level up your AI Skills with Firebase AI and Gemini

Developing AI applications isn't just about writing huge ML code and understanding every LLM. It can also be about integrating simple AI tasks into your apps and becoming 10% more efficient. In this session we'll play with a sample app that extracts different kinds of details from documents, and see how Firebase AI Logic can build simple AI tasks that do the work for us. We'll also touch on AI security and Firebase App Check to protect AI apps from fraud. Code along and build a production-ready AI application.

The talk

The Agent Worked. The System Didn’t.

It's easy to assume that if an AI picks the right tool and generates the correct arguments, everything is fine. But what happens when a user says "Cancel my order and refund me", the cancellation succeeds, and the refund times out? Or a network drop makes the agent retry, two requests cancel the same order at once, or the server crashes mid-workflow? Hemant builds a small AI-powered order operations agent live and pushes it off the happy path, showing how retries, timeouts, partial failures and race conditions turn a smart-looking agent into a data-integrity nightmare. Then the system gets hardened with idempotency, explicit state machines, durable workflows and compensation strategies.

1. 01The illusion of correctness
2. 02Breaking the illusion
3. 03The distributed-systems trap
4. 04Building the guardrails
5. 05Live demo
6. 06Designing resilient agentic systems

The talk

Beyond Vibe Coding: How We Built 20+ AI Agents for a Real-World Startup

Vibe coding can get an AI prototype built in hours. But what happens when that prototype becomes a real product, with thousands of users, multiple AI agents, growing API costs and production reliability challenges? This is the engineering journey behind OneRoadmap, where an architecture supports 20+ specialised AI agents on Gemini: agent architecture, structured outputs, model orchestration, secure execution, failure handling, cost optimisation and deterministic validation. A technical, experience-driven session with real architectural decisions and mistakes, not a product pitch.

1. 01The vibe coding reality check
2. 02Building 20+ specialised AI agents
3. 03Production AI: what nobody shows in demos
4. 04From prototype to scalable architecture
5. 05The production AI blueprint