Buenos Aires • Argentina
LLM Native
LLMNative BA V1 @AWS
Come meet, mingle, and tackle real-world technical challenges, with a special focus on meaningful, production-grade experiences.
Schedule
🏠 Reception
Building LLM-Native Organizations: FieldData's approach
This talk explores how LLMs represent the next automation wave, freeing workers from digital busywork. I'll share the core principles of the LLM-native community, where cognitive tools handle routine tasks invisibly while enhancing human potential. Using Fielddata as a case study, I'll demonstrate practical implementation strategies, internal best practices we use and how we think and develop LLM-powered UX's. Together we'll explore how to build organizations where technology adapts to humans and not the reverse.
Upgrading to Multi-Model: 10x'ing Dev Speed and LLM Quality with Not Diamond
Every week a new frontier AI model is released to the world. And yet the vast majority of developers are still building AI with the 'traditional' approach: inefficiently waterfalling development through one model at a time. This results in inefficient engineering cycles, degraded output quality, and bloated inference costs and latency. In this talk, you'll learn a set of scalable principles and practical data-driven techniques that leading startups and Fortune 100s are using to evaluate and optimize all candidate models and prompt configurations in parallel, saving tens of thousands of engineering hours and improving output accuracy by up to 35%.
Tomás Hernando Kofman
Co-founder & CEO
@Not Diamond
Break
Building Firecrawl's Deep Research API
From launching an open-source version just hours after OpenAI's announcement to introducing the Firecrawl Deep Research API, we’ll explore how Firecrawl drives innovation in the AI era of web extraction.
Nicolas Camara
Co-founder & CTO
@Firecrawl
The Future is Agentic: How Windsurf Writes 90% of its Users' Code
Agentic AI is reshaping software development, shifting AI from an assistive tool to an active collaborator. At Windsurf, we’ve built an integrated coding agent that writes 90% of its users’ code, evolving alongside developers as they build. This talk explores the impact of deeply embedded AI agents—how they accelerate iteration, reduce cognitive load, and change the way engineers approach problem-solving. Drawing from real-world use cases, we’ll discuss the trade-offs of designing agentic systems, the challenges of ensuring reliability in production, and what we’ve learned from features that didn’t make the cut.
Akshat Agrawal
Product + GTM
@Windsurf
Finger food & Networking
Building Agents with Model Context Protocol (MCP)
We'll share the problems we faced in the pre-MCP world and demonstrate how this protocol now unifies our fragmented ecosystem and is revolutionizing agent development. Through practical experiences, we'll show how we've built servers connecting seamlessly with Claude and Windsurf clients, and designed specialized clients for agent frameworks, enabling us to focus on creating sophisticated AI solutions.
Using LLMs to extract tables from Excel spreadsheets
Spreadsheets are ubiquitous in data workflows but contain unstructured, visually formatted tables that pose challenges for automated processing. In this keynote, I’ll show how how to extract meaningful tables from complex spreadsheets with multiple tables, inconsistent formatting, and high-dimensional data. We will discuss our approach at BRINTA, including heuristic-driven metadata generation, tool-assisted LLM navigation, and the crucial role of human validation.
Facundo Palavecino
Sr AI Engineer
@Brinta
How Fairway Applies MMLLMs to Navigate Regulatory Compliance in US Automotive Transactions
This talk will explore how a combination of MMLLMs enables rapid and accurate decision-making in the complex landscape of US regulatory vehicle transactions. We will break down the workflow into key stages: text extraction, signature and notarization identification, transaction categorization, regulatory compliance assessment, and final decision-making—all executed in minutes.
Conrado Marturet
Co-founder & CTO
@Fairway
Benchmarking Localized Contextual Understanding in LLMs through Regional Literature
DECUBO is a benchmark designed to assess the ability of LLMs to interpret literary texts deeply rooted in the cultural, geographical, and linguistic context of Argentina—particularly the Province of Buenos Aires. By leveraging a representative corpus and a curated set of expert-generated questions from the Provincial Directorate of Libraries and Reading Promotion, DECUBO provides a framework to evaluate the limitations and strengths of LLMs in understanding local references.
How Evaluations Improved Darwin's AI Workers Suite
In this talk I’ll share our experience using evaluations to refine our approach to model assessment. We’ll dive into how structured evaluations helped us systematically improve accuracy, reduce hallucinations, and enhance overall performance across our AI Workers Suite.
Ezequiel Sculli
Founder & CTO/CPO
@Darwin AI
Open Networking & Closing
Final networking session and event closing.