When generative AI initially emerged, the demands on Grammarly’s infrastructure grew significantly — but Andriy Derevyanko knew what to do.
The director of engineering is no stranger to rapid change, having joined the company seven years ago during its high-growth startup years. So when the need to evolve the company’s infrastructure arose, Derevyanko was ready to tackle what was in store.
“Scaling our cloud platform to keep pace with this transformation has been both a challenge and a privilege,” he said.
Derevyanko currently leads the teams building and maintaining the cloud platforms and core infrastructure that the company’s AI assistant runs on. Since joining the organization, he has witnessed its technology grow to support tens of millions of users while enabling engineers to build and ship faster — an accomplishment that feels both fulfilling and reflective of continued growth.
If there’s one thing that defines being a leader at Grammarly, it’s the need to embrace evolution, whether that’s related to building infrastructure or, in Stefanie Tignor’s case, understanding data.
“Being a leader at Grammarly, I’ve learned that change is a constant in the AI industry,” the head of product data science and engineering said. “We cannot shield our teams from change — nor would we want to.”
Upon joining the company last year, Tignor has guided the data science team’s efforts to more effectively use data to inform product decisions and strategies. She described her team’s culture as “highly impact-driven,” in which data scientists operate more like product owners than researchers, immersing themselves in the product and business strategy while taking accountability for moving company-level product and growth metrics through their work.
At Grammarly, leaders hold the key to their team’s success, ensuring they have the tools and support they need to drive progress. Himanshu Vasishth has learned this firsthand since joining the company after his previous employer, AI-powered collaboration platform Coda, merged with Grammarly earlier this year.
Once the acquisition was finalized, he continued his role as head of core product engineering and now spends his workdays overseeing a team that’s integrating Grammarly and Coda products. Vasishth’s goal is to help grow the business by increasing the revenue and usage of Coda Docs, the component of Coda’s platform that enables teams to collaborate on shared documents.
He believes that succeeding in this endeavor comes down to tried-and-true teamwork — and the culture he helps cultivate on his team.
“We challenge ideas respectfully, support one another, and take ownership of both successes and failures,” Vasishth said.
As Grammarly continues to grow, Derevyanko, Tignor and Vasishth will be busy carving out an AI-driven future for the company, leaning on innovation, collaboration and empowering team cultures to do so.
About Grammarly
Grammarly’s AI assistant is designed to help teams and individuals craft professionally composed emails and documents. The company’s software enables users to generate content from scratch, instantly accept full-paragraph rewrites, receive personalized suggestions, make their writing clearer and more concise, catch grammatical errors, and strike the right tone for their intended audience.
Describe your time at Grammarly so far. What responsibilities have you had throughout that time, and what are your current goals for your team?
Vasishth: I run the engineering team responsible for all aspects of Coda Docs. My goal is to help grow the business by increasing the revenue and usage of Coda Docs. I work on accomplishing that goal by focusing on multiple aspects: First and foremost, build a best-in-class engineering team and collaborative culture that can execute efficiently; Next, work with go-to-market partners to align on the strategic path we should take; And lastly, ensure we have a clear understanding and alignment on the vision and goals both for the long term and the near term. This is true both for the C-level leadership team as well as the team tasked with delivering on these goals.
Tignor: I lead our product data science and engineering team, which sits within the broader data team and engineering organization at Grammarly. This team of data scientists and engineers is responsible for creating a deep and nuanced understanding of every product and feature we build here at Grammarly. On an ongoing basis, we measure product performance, product utility and value, user engagement, and the quality of our AI. Questions my team asks every day include: Is our product healthy? Is it serving the needs of our users? And is it getting better over time? Beyond fostering a broad and consistent understanding of our products, we also use data to help our product team understand where we should innovate next. We look deeply at user feedback and our most important metrics to determine what new features we should prioritize, and then we track the success of those innovations. Great products — products that truly address user needs and delight users for years — are built using great data, and my team provides that data.
I joined Grammarly in June of 2024, so I am approaching my one-year mark. In this time, my focus has been to help the company more effectively use data to inform product decisions and strategies. This means setting up an organizational structure that promotes fast and easy data-driven decision-making. It means fostering greater connection and collaboration between data scientists and UX researchers, product managers, designers, and software engineers. It also involves creating robust metrics to measure success and spot issues while creating an always-on internal “insights engine.”
Derevyanko: I’m part of our engineering organization and currently serve as the engineering director for Grammarly’s infrastructure team in Europe. In this role, I lead the teams building and maintaining the cloud platforms and core infrastructure that our AI-powered writing assistant runs on. I’m also the site lead for Grammarly’s Berlin hub, helping to grow our engineering presence in one of Europe’s largest tech hubs.
I joined Grammarly over seven years ago when the company was a fast-growing startup with around 140 people. In those early days, like it frequently happens in fast-growing organizations, I wore multiple hats — leading efforts across infrastructure, security, IT and other foundational engineering areas. Over time, I honed my focus on our cloud platform, guiding its growth from a small team of just four engineers in 2018 to an organization of 55 that spans two continents. As Grammarly evolved into a full-fledged AI communication assistant, introducing generative AI to enhance writing and ideation, the demands on our infrastructure grew exponentially. One of the most rewarding aspects of my journey has been seeing our infrastructure evolve while enabling engineers to build and ship faster, including rapidly evolving (and changing) AI technology stack on top of conventional tools and systems.
Tell us about an innovative project your team is currently working on. What is its impact, and what challenges has your team overcome throughout it?
Vasishth: The project I’m most excited about is expanding the underlying shard-based architecture we built for sharing individual pages and applying it to all docs. A Coda Doc can have thousands of pages, dozens of core tables, hundreds of ancillary tables, and interconnected views and controls. Once opened, Coda allows users to make any changes, even offline, by downloading the entire document locally. However, this poses challenges for devices with limited memory and processing power — iPhones being a prime example. Due to these constraints, large docs often don’t perform well on such devices. In reality, users interact with only a small portion of a massive document at any given time. Looking ahead, our team is working on dynamically identifying the essential parts needed for most edits and loading only that subset locally. This approach will enable even massive docs to function smoothly on low-memory and low-power devices.
Designing a system to enforce data security while handling real-time mutations and updates was a complex challenge. Our solution involved creating a shard based version of the document — hydrating the full doc then generating a snapshot of only the content on the specific part the user is looking at, and then persisting that as a shard. When a user opens a page, we load the shard relevant for that page. A key challenge here was validating, we don’t start leaking more data into that shard via an unintentional code change. We built tests that validate any code changes against all existing shards to verify that the data on the shard is still the same. These tests, by nature, continue to cover more cases as usage grows.
Tignor: One project my team is currently focused on is integrating our product metrics given Grammarly’s recent acquisition of Coda. Grammarly is the trusted AI assistant for communication and productivity, and Coda is a collaborative all-in-one workspace. The pairing has so much potential for our product vision and roadmap, but right now it requires us to reconcile two totally different systems and product frameworks. We have this immediate need to combine data sources, and we are also pushed to think about the higher-level questions: How do we operationalize the combined value that we want users to get out of Grammarly and Coda when paired together, and what metric should our product teams strive to improve as we build this new combined product vision?
In my opinion, the biggest challenges with this project are similar to any large project: balancing the need for short-term progress and long-term vision while keeping in mind that the landscape is ever-changing. I strongly believe that it’s always important to show quick wins and short-term success. You must quickly resolve at least one or two tightly scoped specific problems to maintain stakeholder trust and keep team morale high. No one likes to toil on a project with no tangible results for months and months. It’s also very easy to get stuck spinning out about all the different edge cases that could arise in the future. Thoughtful and careful planning is critical; no one should operate without a plan. But it’s risky to get stuck in planning mode for too long; taking action more immediately will often reveal what problems actually need to be solved versus the ones that are more theoretical.
Derevyanko: One of our largest ongoing projects is overhauling our cloud orchestration by migrating from a long-standing AWS Elastic Container Service container platform to Kubernetes. This is a significant undertaking given our high-traffic systems, and we’re managing it in a way that minimizes involvement from feature teams to ensure a seamless migration so that end users never notice the change. We opted for Kubernetes due to its scalability and flexibility; Kubernetes is a de facto industry standard and a cloud-agnostic platform that provides a wealth of open-source tools for developers. Its rich ecosystem, which includes intelligent resource management, deployment workflows, monitoring tools and much more, allows our engineers to iterate and enhance the product more easily once everything is migrated to a new foundation. This project will significantly improve the developer experience, leading to faster feature development and even more reliable service. Ultimately, Grammarly’s active users — over 40 million — will benefit from faster updates and new features thanks to the modernized infrastructure.
“Ultimately, Grammarly’s active users —over 40 million — will benefit from faster updates and new features thanks to the modernized infrastructure.”
As mentioned above, one of the biggest challenges was our desire not to take any time away from the feature teams while migrating their services, and we were largely successful. The goal was to keep their focus on their primary goals while we dealt with product infrastructure. I don’t think I can share any secret recipe here, but I do think that having a tiger team and localization approach helped us. We identified the engineers in one location with the most appropriate skills and in-depth knowledge of services in the scope of the migration. We also invited a few engineers from the feature teams to support us during specific migration stages, such as initial design and testing. This helped us close the knowledge gap and iterate quickly by avoiding time-zone challenges.
How would you describe the culture on your team, and how do you help cultivate it as a leader?
Vasishth: Our team thrives on trust, collaboration, and a growth mindset. We challenge ideas respectfully, support one another, and take ownership of both successes and failures. Mistakes are learning moments — not blame games. We value autonomy, celebrate wins both big and small, and make space for connection and fun.
How Vasishth Fosters A Culture of Trust, Collaboration and Growth
- “Hiring and onboarding: I refine interviews to assess real job skills rather than offer standard questions. I also design onboarding to ensure new hires hit the ground running with the right mentors and structured learning.”
- “Autonomy and learning from mistakes: I trust the team to make decisions, even when they differ from mine. Mistakes are learning opportunities — we focus on what to improve, not who to blame.”
- “Culture and connection: Social bonding fuels resilience, whether that’s through coffee walks, team outings, or shared lunches. We also celebrate unseen efforts with the Golden Shovel Award, recognizing those who go above and beyond.”
- “Visibility and trust: Regular ask-me-anything chats help me stay connected and understand team concerns.”
- “Leading by example: I aim to be vulnerable, own my mistakes and practice ‘steelmanning’ ideas I might not instinctively agree with.”
Tignor: The data team’s culture at Grammarly is highly impact-driven; I expect our data scientists to operate like product owners rather than researchers. As a leader, it is important that I set goals for my team that reinforce this culture of ownership and business/product-mindedness. I view it as my responsibility to organize and orient the team so that they can impact the business. I must also set clear expectations that impact is more important than activity or productivity volume. While impact is super important to me, it’s difficult to drive impact in a low-trust and low-empowerment environment. I also strongly believe in supporting team members’ growth and sense of belonging. Our team has a number of rituals to build connection and collaboration, including our Women in Data group, our monthly newsletter to recognize our team’s accomplishments and social events during our bi-annual, full-team on-sites.
Derevyanko: Our team culture is rooted in trust and an ownership mindset. In practice, this means every engineer is entrusted to take full ownership of their projects and make decisions as if they were the owner of that part of the product. We refrain from micromanagement and instead encourage initiative and accountability. As a leader, I view my role as serving the team — I’m here to empower others, eliminate roadblocks and provide guidance and support when needed. This servant-leadership approach ensures everyone feels valued and enabled to do their best work. Since we operate across various locations globally, keeping the team connected is also a priority. We stress open communication and transparency; for example, we host regular sync-ups where we share demos of work in progress and ensure both remote and office-based team members have a voice in discussions. We also invest in regular on-sites and team-building activities to strengthen our interpersonal connections.
“As a leader, I view my role as serving the team — I’m here to empower others, eliminate roadblocks, and provide guidance and support when needed.”
What are some of the most important lessons you’ve learned being a leader at Grammarly, and how do you apply those lessons to your day-to-day work?
Tignor: To have a team that can successfully manage change and perform through uncertainty, you need to have a foundation of trust, strategic transparency, and a culture of empowerment. You don’t have time to help every single individual pivot every time things change; every person on your team needs to feel connected to the deeper strategy and empowered to act in light of these changes. In my day-to-day work, this looks like strong and consistent objectives and key results for my team, planning processes that give power to bottom-up ideas, and opening up to the team about what I do and don’t know. I also want to call out one of the most important lessons I’ve learned over the course of my leadership career, which is just how much authenticity matters. This means showing up as yourself every day as much as possible, and it means deeply understanding your own unique strengths and weaknesses.
How Derevyanko Approaches Leadership at Grammarly
- “Empower and trust your team: I’ve seen that when you give people ownership and trust, they rise to the challenge. In my daily work, I focus on enabling my team’s success by listening to their ideas, giving them autonomy, and clearing obstacles out of their way.”
- “Embrace fast iteration and continuous improvement: We get our best results by iterating rapidly rather than waiting for a perfect solution. I encourage the team to experiment in small increments and learn from each outcome, which keeps us innovative and adaptable.”
- “Stay humble and keep learning: As a leader, I know I don’t have all the answers. I’ve learned to stay open to receiving feedback and hearing new ideas from everyone on the team. I also try to cultivate an environment where it’s safe to admit and learn from mistakes. This mindset helps us solve problems faster and builds mutual respect.”
What are you most excited to tackle with your team over the next year?
Tignor: Over the next year, I’m most excited for our team to really push the boundaries of our product and drive user growth in a major way using data. Grammarly is a well-known and well-loved product for our writing suggestions. I am so proud of that product, and of course, we’ll continue to grow and expand that experience. But we’re also going to be investing heavily in new AI experiences that are highly personalized and contextualized. Doing this effectively requires a strong data science team and a strong product mindset, so my team is well-oriented to build amazing things in this space. I’m so excited to deliver AI to users that is human-centric and meets real user needs.
“I’m so excited to deliver AI to users that is human-centric and meets real user needs.”
Derevyanko: There’s so much to look forward to in the coming year. First, I’m excited to fully capitalize on our new Kubernetes-based platform once the migration is complete. With this modern infrastructure in place, we’ll be able to introduce numerous improvements for our developers, such as faster and more intelligent build and deployment pipelines, more sophisticated monitoring and alerting, and ultimately enabling engineers to focus on what matters most, which is the value we create for our users.
I’m also eager to see our Berlin engineering hub continue to grow. We plan to hire more talented individuals and expand our expertise here, which will further strengthen Grammarly’s global engineering capabilities. Ultimately, what excites me most is the innovation we can drive; by empowering our team with better tools and a culture that embraces rapid iteration, we’ll be well-positioned to build the next generation of Grammarly’s product features and infrastructure enhancements. We have an opportunity to significantly boost Grammarly’s ability to help people do their best work, and that’s a motivating vision for all of us.
“We have an opportunity to significantly boost Grammarly’s ability to help people do their best work, and that’s a motivating vision for all of us.”
I want to emphasize that it’s an incredibly exciting time to be tackling engineering challenges at Grammarly. The problems we’re addressing range from scaling AI-driven systems and building platforms that simplify developers’ lives to creating cutting-edge large language model-based product experiences for our users at a global scale. For anyone who wants to learn more about what our engineering team is working on, I encourage you to explore the Grammarly Engineering Blog for in-depth posts about our projects and innovations.
