Join Kostia Omelianchuk and Lukas Beisteiner as they unpack the full scope of Grammatical Error Correction (GEC) from task framing, evaluation, and training to inference optimization and serving high-performance production systems at Grammarly. They will discuss: The modern GEC recipe (shift from heavily human-annotated corpora to semi-synthetic data generation), LLM-as-a-judge techniques for scalable evaluation, and techniques to make deployment fast and affordable, including Speculative Decoding.

Discussion on removing speed limits for the development of agentic features, from initial LLM-powered features at Grammarly to a modern, first-class agentic surface.

A practical, provocative session full of stories, ideas, and insights to power your AI product journey. Learn how product managers can identify problems worth solving with AI, integrate AI into more holistic solutions, and avoid building shiny features no one needs.

In this talk, we will examine how LLM outputs are evaluated by potential end users versus professional linguist-annotators, as two ways of ensuring alignment with real-world user needs and expectations. We will compare the two approaches, highlight the advantages and recurring pitfalls of user-driven annotation, and share the mitigation techniques we have developed from our own experience.

How can we influence quality during the prompt creation stage, as well as how to work with already-generated text—improving it, identifying errors, and filtering out undesirable results. We'll explore linguistic approaches that help achieve better, more controlled outcomes from LLMs.

Valerie Vegera, Senior UX Researcher at Grammarly, invites you to discover how AI is evolving from a novel curiosity to an indispensable partner in people's daily workflows. The talk explores designing AI experiences grounded in real human needs and how to integrate them into the ways people work.

Two Grammarly data scientists discuss measuring AI ROI from two angles: one builds a novel experimentation framework, the other designs a new scoring system to quantify impact. They cover how each approach was designed, implemented, and validated, and share lessons learned.

Panel featuring Olena Nahorna, Katarzyna Stoltmann, Jennifer Lapp, Aliya Boranbayeva, moderated by Anita Fechner, discussing AI in communication and data.

The talk explores how large language models (LLMs) have accelerated the development of linguistic features. It focuses on how to adapt feature development processes to match this rapid pace and highlights key considerations for maintaining high-quality output in a fast-evolving AI landscape.

Journey of developing an LLM-based product that users like, from cutting through AI hype to setting realistic goals, rapidly building prototypes, and making AI valuable for users.

Andrew Garkavyi and Lesha Levzhynskyi discuss the history and present state of shipping features across Grammarly's multiple platforms, recounting challenges and approaches from fully native to web to hybrid, and addressing overlays, assistant and chat modes in the age of LLMs.

Take a peek behind the curtain at how Grammarly for macOS uses private APIs to replace text in Apple Notes and provide a flawless user experience. We’ll also talk about how to find and use private APIs on Apple platforms.

How we do feature experimentation in the macOS Grammarly app

• The basics: experiments, flags, holdouts, audience filters
• Development and testing
• Checking the metrics: analyzing some real cases
• Benefits and drawbacks
• Real-world examples

Learn what it took for our team to replicate the visuals and interactions of the system keyboard so we could provide users with a smooth transition to our keyboard powered by Grammarly functionality. You will be surprised that it requires a bit more than creating 30 UIButtons. We will discuss different animations and interactions that we had to implement. We’ll also describe how we attained feature parity with the system keyboard regarding emoji, which some other third-party keyboards don’t offer. Finally, we’ll explain why original architecture decisions may not always be bulletproof, even for such a simple thing as a keyboard.

During the session, we’ll discuss the challenges that prompt engineering has presented, both when it first gained popularity and as it continued to evolve. We’ll share how these challenges informed the development of our prompt engineering tooling and workflows. We’ll cover: Standardizing communication with LLMs; Using templating to customize prompts; Building prompt-centric production workflows; Working with structured LLM output; Ensuring the quality of LLM output; Creating tooling that supports our prompt engineering workflows.

How we tackle infrastructure migrations at scale, with hundreds of services and dozens of dev teams; Changing culture to aid teams in using the new tooling; DevX practices and tooling for team onboarding and self-service; Internally developed tools to keep on-call rotation manageable during and beyond the migration process.

Panel with Grammarly Engineers and former interns on our summer internship program and tips to kick off your career.

In this talk, we’ll share how we use LLMs’ power to design intuitive and user-friendly features. Learn about our journey from concept to implementation, the challenges we face along the way, and our practical solutions.

LLMs have unlocked new opportunities in NLP with their possible applications. Features that used to take months to be planned and developed now require a day to be prototyped. But how can we make sure that a successful prototype will turn into a high-quality feature useful for millions of customers? In this talk, we will explore real examples of the challenges that arise when ensuring the quality of LLM outputs and how we address them at Grammarly.

Overview of Grammarly's in-house solution for conducting quality evaluations of suggestions, including what human quality evaluations are, how the solution provides insights into the impact and quality of new features before deployment, and a deep dive into the scalable, distributed design of the solution.

LLMs have opened up new avenues in NLP with their possible applications, but evaluating their output introduces a new set of challenges. In this talk, we discuss how the evaluation of LLMs differs from the evaluation of classic ML-based solutions and how we tackle the challenges.

LLMs have opened up new avenues in NLP with their possible applications, but evaluating their output introduces a new set of challenges. In this talk, we discuss these challenges and our approaches to measuring the model output quality. We will talk about the existing evaluation methods and their pros and cons and then take a closer look at their application in a practical case study.

This talk covers Grammarly's approach to using a combination of third-party LLM APIs and in-house LLMs, the role of LLMs in Grammarly's product offerings, an overview of the tools and processes used in our ML infrastructure, and how we address challenges such as access, cost control, and load testing of LLMs, sharing our experience in optimizing and serving LLMs.

🚀 Writing efficient unit tests is essential for critical software components. But how do we ensure that our tests are thorough enough to cover a huge variety of typical inputs and most edge cases? I'll introduce you to the concept of property-based testing and share how we benefited from using it at Grammarly.