Kaustubh (@_ofkaus) and Prashant (@primaprashant) sit down with Hazumu Yamazaki (@HazumuY), co-founder and co-CEO of Empath, to talk about challenges in building an AI product - from getting initial users to serving 3100 customers in 50 countries, from starting with zero data of their own to creating their data supply chain, and more.
Do you love our content and want to help us grow? Leave a review on our website.
Want to be among the first to hear from us? Follow us on Twitter @techculturepod.
- 00:00 - Intro
- 01:17 - What's Empath all about?
- 02:31 - How Hazumu's PhD research led to Empath
- 04:37 - Market research and validation for emotion recognition API
- 07:47 - Getting initial users and longer B2B sales cycle
- 09:55 - Founding team
- 11:11 - Becoming sustainable
- 12:16 - Improving productivity of online meetings with JamRoll
- 16:55 - Interesting feature requests like lie detection
- 18:15 - Problems with emotion recognition in videos
- 21:48 - Collecting data for a new AI startup
- 23:47 - Adding labels to the collected data
- 24:45 - Recognizing emotions in different languages
- 26:05 - Sarcasm and hidden contexts in conversations
- 27:49 - What distinguishes Empath from publicly available models
- 30:06 - Tech stack at Empath
- 32:02 - Adoption of AI Japan
- 33:40 - Advice for the initial stages of building an AI product
- 35:11 - Organizations that support AI startups in Japan
- 37:10 - Going global
- 38:54 - How to make a great pitch
- 41:23 - Funding rounds
- 42:00 - Biggest challenge in reaching the next milestone
- 43:46 - Long term vision for Empath