AI Conf 2026

Talk details

What does it take to build a production-ready AI system that processes 120,000+ complex documents at around $2 per document? This talk shares lessons from building an AI engine that digitizes and structures unstructured clinical trial data at enterprise scale.

We'll explore how the team preserved layout and document structure, detected and extracted tricky tables (nested, merged cells, domain-specific formatting), and used chain-of-thought reasoning with citation tracking to keep outputs verifiable in a regulated environment.

You'll also hear the operational realities of running high-throughput inference on AWS Bedrock, balancing RPM/TPM, accuracy vs. speed trade-offs, and choosing between custom and foundation models after spending over half a million dollars on research and infrastructure.

Talk details

AI is redefining the role of designers - shifting them from hands-on execution toward leadership, strategy, and decision-making at scale. As AI tools rapidly automate production tasks, the true value of designers increasingly lies in direction, clarity, and creative judgment.

In this session, Dijup Tuladhar explores how tools like Figma, Figma Make, and Copilot are transforming design workflows by accelerating research synthesis, enabling rapid exploration of ideas, and streamlining design operations. The talk focuses on how AI supports faster, more confident decisions and strengthens collaboration across product, engineering, and business teams.

Through practical examples and real-world workflows, attendees will learn how senior and principal-level designers can leverage AI to scale their impact without compromising quality or vision. Rather than replacing designers, AI becomes a force multiplier - freeing them to think more strategically, lead more effectively, and shape better product outcomes.

Talk details

We'll explore how to build LLM systems worth trusting and designed with intention.

When LLMs first appeared, excitement centred on capability. But as systems scaled, new challenges emerged: trust, transparency, and reproducibility became harder to maintain.

This session bridges art and science: the creative problem-solving needed to shape LLM behavior with the engineering discipline to make it reliable.

You'll learn mental models for building systems that are secure by design, explainable by default, and governed from the start. We'll cover how thoughtful design enforces guardrails, surfaces model understanding through telemetry, and keeps humans engaged at decision points.

Talk details

Briefly speaking, let's look at the current landscape of programming languages and imagine what a next-generation language designed for both humans and machines would look like.

The software development landscape is undergoing a seismic shift. Three years after ChatGPT's debut, AI coding assistants have moved from novelty to necessity. But here's the paradox: while AI can write code in any language, the choice of language dramatically affects the quality, cost, and reliability of AI-generated code. This talk raises a fundamental question: What characteristics define the perfect programming language for the AI era?

In this talk, we'll explore the evolution of programming languages in the age of AI: how expressiveness, safety, and performance converge to form a new paradigm of communication between humans and intelligent systems. We'll compare modern statically and dynamically typed languages such as Rust, Python, Go, and Ruby, and draw lessons from their strengths and limitations. We'll discuss the economics of "cheap tokens" versus "expensive correctness," how compilers become feedback systems for agents, and why traits, macros, and extensibility matter more than ever.

Talk details

Everyone talks about chatbots, but what capability actually changed the world? This talk contrasts old AI limits with what became possible once LLMs could turn messy input into structured decisions, actions, and content.

We'll explore how LLMs reason, plan, and drive workflows, and why hallucination is a consequence of how these models generate ideas—not just a bug.

Through practical examples, Manu reveals the real "there there" behind GenAI and how intent-to-output systems reshape how builders design products and architectures.

Talk details

"LLM Evaluations + Observability (Evalobility)" explores how modern AI teams align on a single dashboard for evaluations, experiments, and monitoring.

The session walks through how designers, PMs, and engineers use shared traces to keep LLM outputs helpful, including Datadog LLM Observability screenshots and trace examples tailored to these workflows.

You'll see practical ways to balance performance, cost, and output quality by tracing model behavior, validating UX impact, and keeping human feedback in the loop.

Talk details

AI security must be considered from multiple angles - preventing users from exploiting your AI platform, making sure your AI does not produce harmful content, and keeping personal information of end users safe.

"Non-deterministic automation" is taking over almost everything and new threat surfaces are emerging faster than we can keep up.

The creation of environments to facilitate end users' AI agents taking actions on their behalf is also quite interesting from a security and liabilities perspective.

Tool call poisoning, retrieval of "private" info that leaked into the training data, and training data poisoning are also topics I've planned to speak on.

My presentation will mostly revolve around these ideas.

Talk details

Few would argue against the notion that AI will have a transformative effect on humanity, likely within our lifetimes. Yet we still as of yet have a very poor understanding of how AI works at a fundamental, mathematical level, and thus can't control it nor predict its behavior. Moreover, individuals and institutions alike seem to be rushing to adopt AI in spite of this fact, which for me is a cause for grave concern.

My talk will be on AI for humanity, AI safety for everyone (especially children), and encouraging all the AI builders and engineers in attendance to slow down and think more deeply about the human implication of their work. I will also talk about my own research at the intersection of AI, neuroscience, and education.