Sujatha Sagiraju, Chief Product Officer at Appen – Interview Series
Sujatha Sagiraju is the Chief Product Officer at Appen, she joined Appen in September 2021 as SVP of Product and she is responsible for the product strategy. She is a technology pioneer with over 20 years of broad experience in building disruptive large-scale online services and AI/ML and data platforms. She joined Appen from Microsoft where she held leadership roles in several groups including Bing and Azure AI Platform.
Appen is the global leader in data for the AI Lifecycle. With over 25 years of experience in data sourcing, data annotation, and model evaluation by humans, they enable organizations to launch the world’s most innovative artificial intelligence systems.
What initially attracted you to AI?
When I was at Microsoft, I worked in Azure AI organization. I was familiar with the industry landscape, the customers, and the AI transformation that is happening across different industries. I could see from a customer’s point of view that training data was a roadblock to building machine learning models and I saw Appen as an opportunity to solve that problem – the missing link that could connect all the stages of the AI lifecycle.
You’re currently the Chief Product Officer at Appen, could you describe what this position entails?
At the highest level, my team builds the product vision, strategy, and aligns with multiple different stakeholders across the organization in effectively executing it. On a more personal level, I spend considerable time understanding the industry and customers. With some of the largest companies as our customers such as Amazon, Google, Microsoft, Salesforce, Boeing, it’s important for my team to understand the customer scenarios and pain points and build a product strategy that delivers a growth plan. Building a safe, inclusive culture is also a very big part of my role as I focus on creating a space for our employees to share ideas, collaborate, and grow their careers.
How important to AI development is fostering diverse teams?
It is extremely important for AI development to have diverse teams. There are a few different ways to think of diversity – gender, age, race, perspectives. The diversity of perspectives can be the most important part of making sure you have diverse backgrounds and experiences on your team. Those experiences bring new and different ideas to help build the best product for all your customers that are very diverse.
How do you create a work culture that synergizes this diversity?
A culture that promotes diversity invites employees to share their ideas and perspectives. I like to consider different methods of communication when conducting team meetings. For example, when asking for feedback in a team meeting, I ask for employees to speak directly in the meeting or send me a message after they have thought it over. I recognize that not everyone would like to speak or share feedback right away, and I want to create a culture where that is acceptable. I want a safe environment for people to voice their opinions and share their ideas however they prefer. Great ideas come from all different teams within the organization. I meet with sales, marketing and other customer facing teams to understand their needs with the product and their perspective working closely with customers. Some of the best product ideas come from listening intently to the pain points of the customers – either directly from them or teams that interact with our customers each day.
Outside of having diversified teams, what are other ways of fighting bias in machine learning algorithms?
Inclusive data sourcing, data preparation, and model evaluation are critical to fighting bias. The data used to train the algorithms must be inclusive of all potential end-users or outcomes. When moving through different stages of the AI lifecycle, each stage must be checked for bias. Responsible AI is also built with responsibly sourced datasets meaning the contributors are treated fairly. Appen built a Crowd Code of Ethics to show our dedication to the well-being of our Crowd.
You recently posted an article discussing a new discipline, called Data for AI Lifecycle. Could you briefly describe what this is?
The Data for the AI lifecycle encompasses four stages in a continuous cycle; data sourcing, data preparation, model build and deployment, and model evaluation by humans. These stages are necessary to deliver high-quality data for building AI projects. Data sourcing, data preparation, and model evaluation are the most laborious and data-intense and if not done well, can lead to project quality issues and launch delays. Appen specializes in those three stages and strategically partner with providers who specialize in model training and deployment.
What’s the role of synthetic data in the Data for AI lifecycle?
Data sourcing solutions include human-annotated data, pre-labeled datasets, and synthetic data. Synthetic data is leveraged in hard-to-find datasets and use cases. Inclusive datasets cover all use cases and potential end-users of an AI model, and some require synthetic data to reach that goal. The combination of human-annotated data and synthetic data will become critical to model success.
How big of an issue is model drift or overfitting with the Data for AI lifecycle?
Model drift can be a big issue and needs to be addressed in the fourth stage of the AI lifecycle, Model Evaluation by Humans. It’s critical that the model continues to work in the real world and to know that it must go through human testing. As environments change and grow, models need to change as well. It’s important that practitioners continually evaluate their models to prevent them from becoming outdated or biased. Microsoft’s Bing is a customer who utilizes model evaluation to ensure search results are performing to their standard and the model is continually being evaluated.
Is there anything else that you would like to share about your work at Appen?
The most valuable work at Appen is by our people and their expertise. With 25 years of experience, Appen has built a strong foundation with its employees. Our customers trust our expertise to deliver high-quality results, quickly and at scale. Appen is enabling the AI industry transformation by providing solutions to seamlessly manage the Data for the AI lifecycle.
Thank you for the great interview, readers who wish to learn more should visit Appen.