AI workflows and teaming - how do teams deliver AI products?
summary
Panel argues a three‑stage checklist: discovery → prototype → production with concrete short tests (quick prototypes) then full MLOps (deploy, SLIs/SLOs, monitoring, rollback). Common failures come from handoffs between researchers and engineers, lack of model testing, and missing explainability; one cited outcome is an NLP system for regulatory docs that survives internal use only after operationalization. Investment priorities are engineering discipline, data‑centric ML and synthetic data generation, plus clear team design versus chasing a single “unicorn” role, i.e. ML engineering discipline.
AI-written summary · about
chapters (8)
- 0:00Start
- 1:35Ashley Beatty
- 5:37Why Ai Products Are So Messy
- 18:29Product Discovery
- 22:21What Are the Remaining Challenges
- 30:28When Was the Last Time that You Were Truly Amazed by Ai
- 35:59What Is the Role of Synthetic Data in Your Applications and Are Your Teams Interested in Investing in Building a Resources To Develop More Synthetic Data
- 36:42Data Centric Machine Learning
related talks
33mAI with a DevOps mindset - experimentation, sharing ..
54mArtificial Intelligence seen from the software development lifecycle perspective
13mMLOps: From AI Research to Production-Ready Systems
26mAre you ready for MLOps? 🫵
30mBuilding Teams and Culture that Support ML Innovation
22mHow to operationalize AI
56mPanel: Scaling AI successfully
43mNavigating the MLOps Journey: Key Considerations for Successful Implementations