Gareth Jones, data scientist at Arturo, recently spoke with our COO, Brian Rieger on how Arturo’s machine learning team built mature and complex AI-first products while juggling resource costs, data labeling and scalability as top priorities.
Headquartered in Chicago, Arturo is a deep-learning company that focuses on providing clients with AI property insights for residential and commercial projects.
As an ML practitioner, Gareth offered a few key lessons learned and what data science teams tackling similar problems can learn on their path to ML production.
You can check out the full recording here!
Specific questions topics they discussed include:
- How do you size up a problem to ensure there is a higher chance of a successful ML modeling outcome?
- How does high-quality training data fit into this equation?
- What are some of the critical insights when thinking about building AI applications as it relates to data volume, variety and velocity?
- What are some best practices for managing growing software complexity with growing labeling complexity?
- The interplay between the different roles like labeling ops, operations, ML engineer and taking pride in your data as a team.
- The importance of having a training data platform when you’re looking to expand your AI application to new territories and use cases.
- How and why the last thing a sophisticated ML team wants to do is build their own labeling tools.
- Irrespective of industry, how can more companies succeed with AI and what are some key trends emerging that facilitate building these types of applications?