Podcast
Turn Data Science Project Failures into Career Wins: Production Lessons, MLOps Fixes & Framing Failures on LinkedIn
Open original DataTalks.Club episode
Turn Data Science Project Failures into Career Wins: Production Lessons, MLOps Fixes & Framing Failures on LinkedIn
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 turn data science project failures into tangible career wins — and how should you talk about them on LinkedIn? In this episode, Yury Kashnitsky, Ph.D. in applied math, Kaggle Master and Senior ML Scientist at Elsevier who also leads the open course mlcourse.ai, walks through real production ML lessons and MLOps fixes learned across academia, startups and industry.
People
Use these links to connect the episode to guest notes.
Chapter Summary
Use these checkpoints to decide whether to open the source transcript.
- 0:00 - Episode Introduction
- 1:30 - Episode Theme: Failures and LinkedIn Omissions
- 2:32 - Guest Opening: Background Snapshot
- 3:05 - Career Journey: Aviation, Academia, and Transition to NLP
- 4:58 - CV Choices: Omitting Hobbies and Personal Details
- 5:35 - Project Failures Overview: Common Data Science Pitfalls
- 6:22 - Case Study — Proofreading AI: BERT Regression and Early Termination
- 11:06 - Stakeholder Communication: Making the Call to Stop a Project
- 11:31 - Product Management Gap: Value of a Data Product Manager
- 16:46 - Customer Development: Rapid Validation vs Building ML First
- 18:00 - Engineering vs Research: Deployment and Serving Constraints
- 19:04 - Production Lesson: Gradient Boosting vs CTR Heuristic Baseline
- 25:25 - Performance Fix: Re-ranking Scope Reduction to Meet Latency
- 25:56 - DevOps Anti-patterns: SSH Deploys, No CI/CD and Technical Debt
- 28:11 - From Notebooks to Production: BI, LTV Predictions, and MLOps Needs
- 30:44 - Startup Anecdote: GPU Overstock, Bitcoin, and Sentiment Analysis
- 34:36 - Data Labeling Reality: Cost, Quality, and Mechanical Turk
- 35:18 - Resume Strategy: Omitting Short or Sensitive Startup Stints
- 36:12 - Telco NLP: Multilingual Complaint Classification & Transfer Learning
- 39:54 - Too Much Freedom: Research Time vs Impactful Production Work
- 41:07 - Interview Tip: Ask About Active Revenue-Producing ML in Production
- 43:20 - Digital Presence: GitHub, Open Courses, Talks and Hiring Impact
- 45:35 - Work-Life Balance Hacks: Focus Time and Side Projects
- 48:27 - Public Activity ROI: A/B Tests, Talks, and Career Opportunities
- 49:30 - Framing Failed Projects on LinkedIn: Honesty and Lessons Learned
- 52:41 - Business-Travel Boundaries: Perm Trips and Weekend Work Limits
- 58:36 - Closing Thoughts: Embracing Failures and Building Resilience
- 1:00:24 - Contact & Resources: Open Course and Social Links