Machine Learning Archives - Software Engineering Daily BaseTen: Creating Machine Learning APIs with Tuhin Srivastava and Amir Haghighat
May 20, 2021
Tuhin Srivastava, founder of BaseTen and former data scientist, joins forces with Amir Haghighat, co-founder and expert in backend systems, to discuss the evolution of machine learning APIs. They delve into the challenges of integrating model outputs into workflows and simplifying ML deployment for data scientists. Tuhin shares insights on deploying a style-transfer model and a fraud operations pipeline, while Amir highlights BaseTen's serverless architecture and user-friendly UI builder. Their vision aims to lower the barrier for internal ML adoption across industries.
AI Snips
Chapters
Transcript
Episode notes
Models Are Only Useful When Integrated
- A model is merely an artifact that maps inputs to outputs and needs serving to be useful.
- Real production complexity is integrating outputs into business workflows and feedback loops.
Let Data Scientists Ship Glue Code
- Deploy models behind APIs and let data scientists write glue code in Python rather than forcing them to build web stacks.
- Provide tools for versioning, chaining, and UI so models can be turned into user-facing applications quickly.
Style Transfer Demo Became A Usable App
- Tuhin described deploying a style-transfer model and immediately creating a drag-and-drop UI so nontechnical users can try it.
- BaseTen lets you deploy the model, wrap Python business logic, and build a frontend without writing Flask or React.
