Podcast
From Research to Production: Build Reproducible, Deployable Full-Stack ML Systems
Open original DataTalks.Club episode
From Research to Production: Build Reproducible, Deployable Full-Stack ML Systems
Original Episode
Use these links for the canonical episode and media sources.
- Open the original DataTalks.Club podcast page
- Watch on YouTube
- Listen on Spotify
- Listen on Apple Podcasts
Episode Overview
How do you move ML work from research notebooks to reproducible, deployable full-stack systems? In this episode, Mihail Eric — founder of Pametan Data Innovation and Confetti.ai, former Stanford NLP researcher with industry experience at RideOS and Amazon Alexa, and author of papers in ACL, AAAI, and NeurIPS — tackles that exact challenge. We trace Mihail’s path from academic NLP to self-driving and conversational AI, then into hybrid roles that blend hypothesis-driven research with production engineering.
People
Use these links to connect the episode to guest notes.
Chapter Summary
Use these checkpoints to decide whether to open the source transcript.
- 1:17 - Podcast Introduction
- 1:52 - Guest Overview: Mihail’s Roles and Work
- 2:00 - Guest Background: Stanford NLP and Early Research
- 5:00 - From NLP to Self-Driving: Shared Long-Tail Challenges
- 6:46 - Transition to Industry: Building Engineering Foundations
- 8:34 - Research Infrastructure: Data Collection and Prototyping
- 9:21 - Hybrid Role at Amazon: Research Integrated with Production
- 10:52 - Researcher Focus: Hypothesis-Driven Work and Benchmarks
- 12:50 - Experimental Tooling: Notebooks, W&B, Fast Prototyping
- 14:45 - Sourcing Research Questions: Surveys, Citations, and “Future Work”
- 17:35 - ML Engineer Focus: Full ML Lifecycle and Production Systems
- 17:53 - Engineering Tooling: PyTorch, Docker, Cloud, and Web Frameworks
- 20:25 - Data Science Evolution: From Data Science 1.0 to Data Science 2.0
- 23:32 - Skills Swap — Researchers Learn: Engineering Rigor and Reproducibility
- 28:50 - Skills Swap — Engineers Learn: Handling Uncertainty and Experimental Rigor
- 30:16 - Bridging the Gap: Cultural and Organizational Challenges
- 34:20 - Embedded Teams vs. Handoffs: Avoiding the “Throw-It-Over-the-Wall” Trap
- 36:57 - Breaking Silos: Leadership, Sprints, and Active Collaboration
- 39:08 - Role Fluidity: Flexible Responsibilities in High-Performing Teams
- 40:33 - Full-Stack Data Scientist: From Model Development to Deployment
- 44:36 - Advice for Researchers: Build End-to-End Systems and Deploy
- 46:57 - Code Reviews for Researchers: Rapid Engineering Skill Development
- 47:51 - Advice for Engineers: Read Papers, Reproduce Models, Run Experiments
- 51:28 - Practical Paper Reading: Tutorials, Code, and Researcher Collaboration
- 55:31 - Choosing a Path: Internships, Masters, PhD — Try Both Early
- 58:56 - Confetti.ai: Career Preparation and Learning Resources for ML Roles
- 1:01:40 - Contact & Resources: Twitter, LinkedIn, and Confetti.ai