Tokyo startup Baobab solves for training data for machine learning
How a woman founder is driving data construction—and economic opportunity
Japanese founder Miori Sagara was working with machine learning for language translations when it was just being integrated into most technologies. Collaborating with government offices and academics, she developed a corpus-based machine translation for the e-commerce industry. Although the project was cut short, the potential of combination of human skill and machine learning was still on her mind. So three years later Miori founded Baobab, a startup designed to provide annotation data for machine learning. The company's platform uses machine learning to create translation engines specialized for particular areas, focusing on providing improvements in efficiency and image annotation, tagging, and captioning of images and videos, and voice transcriptions. It then builds and provides data to meet the diverse and often highly specialized needs of corporations and research institutions in Japan and overseas.
Baobab uses human annotators, nicknamed “Baoparts,” to help with their machine learning data. Hailing from 22+ countries all over the world, Baoparts vary widely in their careers and experience; many live in rural areas or islands where employment is scarce, some are stay-at-home parents hoping to secure a secondary income, and many self-identify as living with developmental disabilities like autism. Miori takes pride in the diversity of her Baoparts, especially when it comes to hiring the people with disabilities, so she was keen on learning how to create an even more sustainable and welcoming workplace for all aspiring Baoparts. One Baopart from Cambodia left the feedback: “I appreciate how Baobab always treats us as a true partner, never as an item on a list.”
Since launching in 2010, the scale of Baobab’s clients has grown both in size and in geography, expanding to overseas markets. In parallel, their Baoparts were becoming even more multinational and diverse. As the company was entering this new stage of growth, Miori wanted to expand in a more sustainable way. While watching the news, Miori found out that Google for Startups was launching an accelerator program in Tokyo. “Google offers services to the entire world, and I wanted to learn how Baobab could expand our services on a global scale.” says Miori. So, she enrolled in Google for Startups Accelerator: Japan, a three-month program that consisted of in-person training and mentoring sessions at the Google for Startups Campus Tokyo space.
“Google mentors treated me without preconceptions so I was able to stretch myself. I felt confident in being myself at every moment of the program.”
Miori took part in a one-on-one Google UI/UX session where she learned tips on how to better communicate with her remote workforce. On the advice of a Google mentor, she also explored the Google re:Work website’s collections of best practices, research, and ideas designed to help managers put people first. Miori realized that given the global nature of Baobab, it’s important to give more user-friendly guidance to her employees. Miori decided to add visual aids to their guidelines to make it more accessible for all of their Baoparts worldwide. An instructor at one of Baobab’s facilities notes that the changes in the environment has dramatically increased the mental health of the workers - particularly those with autism. He adds that there is now camaraderie between staff where there was silence before and absences have gone down dramatically. Miori added, “Even after a year since the program started, we are still continuing to implement and build off of advice I received from Google mentors.”
After graduating from the Google for Startups: Accelerator Japan program, Baobab received their first external funding from an angel investor in November 2020 and another funding from the KIBOW Impact Investment Fund in March 2021. "As a woman founder, I had always been afraid that everything would go more smoothly if I didn't express my opinion,” said Miori. “Google mentors treated me without preconceptions so I was able to stretch myself. I felt confident in being myself at every moment of the program.”
Baobab’s technology is fueled by Google products. The platform uses TensorFlow and Cloud’s Auto ML to create models and quickly analyze large amounts of data to provide clients with models and benchmarks, before they add the final customizations and complete the project. In the near future, Baobab will not only build training data for image annotation, but also release a model evaluation service using Google's AutoML. “Our vision is to power the future by creating a society where people with real problems can take advantage of AI and create a world where anyone can build an AI model easily anywhere – just like building Legos.”