AI

Sessions

Artificial Intelligence

Designing Conversations: The Art and Science of Building Voicebots

In this session, Ruchir shares a journey of building a Minimum Viable Product (MVP) for a voicebot solution. From selecting the right tech stack and handling speech-to-text challenges to designing conversational flows and ensuring platform integration, this talk covers the practical decisions, hurdles, and key learnings from our MVP development.

10:30

11:00

A Recipe for Real Time Voice AI: WebRTC, SLMs and Open Source Software

This talk explores how Small Language Models (SLMs) and open-source tools enable affordable, fast AI-driven WebRTC interactions.Based on real-world experience building AI analytics and real-time voice agents, we’ll cover: 1.How SLMs reduce cost and improve response times for voice AI. 2.Some standard AI architectures that connect models to external data and tools. 3.Benchmarking some voice AI pipelines to compare real-time performance. 4.Live demo of an AI-powered WebRTC assistant.

11:45

12:15

5 Lessons From Deploying Voice AI in Production BPO Environments

Building a voice AI agent is one thing, but listening to real customers interact with it can surprise you. In this talk, Justin shares five lessons learned deploying production voice AI for BPOs, including what happens when function calls fail, callers leave early, or flows become unexpectedly complex. Dive into strategies for regression testing, prompt engineering, and structuring multi-step flows that perform reliably in the real world. You’ll also learn how we designed smooth handoffs to human agents when voice AI reaches its limit.

2:45

3:15

Beyond IVR: Why Conversational Apps are The Platonic Ideal of SaaS and e-Commerce

The platonic ideal of a B2B website: give the visitor everything they need to fulfill their intent without talking to a human. The most sophisticated B2B and ecommerce brands go to great lengths to reach for this platonic ideal. Almost all of them eventually need humans in the mix to answer pre-sales questions, facilitate complex sales and implementations, or deliver answers to routine and complex support questions. If LLMs continue to evolve at their current pace, Generative IVR (a cute name for conversational apps that use a mix of generative AI and IVR-like deterministic workflows) offers the perfect framework to deliver on websites' platonic ideal: a self-service system for fulfilling every stage of the customer realtionship minimal involvement from humans.

3:15

3:45

XML For the Win: Reliable AI Data Extraction

Getting deterministic(ish) results from non-deterministic generative AI models can be a challenge. Stop begging or threatening AI to conform to your desired output format, and surrender to the comfort of XML templates. In this short talk, I'll outline a simple approach to reliably extract structured data from a wide range of generative text models -- one approach to rule them all -- as well as extra tips to build battle-tested pipelines that integrate AI.

3:45

4:15

AI Workshop ft. Brian West

More details coming soon!

9:30am

10:15am