From notebook to production using

on YouTube ↗01h 00m

" summary

Code converts a notebook into a Flask API, a React frontend and a live Heroku deployment, with the notebook refactored into domain models, repository interfaces and tests. A slow kernel density / nearest‑neighbor approach gets replaced by a compiled Hungarian algorithm implementation, cutting memory from ~20–30MB to about 5–10MB. Data lives in s3 or local pickle repos, results serialize to JSON, search scores return frame ids and the prototype UI and API both work end to end.

AI-written summary · about

" chapters (16)

1. 0:00Notebook to production
2. 2:57Video analysis problem
3. 8:03Domain models from notebook
4. 12:23Hungarian algorithm optimization
5. 16:51Unit tests for matcher
6. 20:04App and infrastructure layers
7. 26:23Tracking data visualization
8. 31:07Flask API for search
9. 35:30React frontend visualization
10. 40:17Heroku deployment
11. 46:45Cloppy library intro
12. 51:28Open source resources
13. 52:09Q&A memory optimization
14. 53:56Q&A response time scaling
15. 56:11Q&A frame filtering
16. 58:00Q&A open source contributions

" related talks

• 27mNotebook To ProductionNir Barazida · PyData Tel Aviv 2022
• 34mNotebooker: Production and Scheduling for your Jupyter NotebooksJon Bannister · PyData London 2022
• 20mFrom notebooks to production - bringing devops to scientistsRaam Rosh Hai · GeeCON 2019
• 49mUntitled13.ipynbVincent D Warmerdam · PyData Amsterdam 2025
• 40mPython and pandas as back end to real-time data driven applicationsGiovanni Lanzani · PyData Berlin 2014
• 01h 23mImproving Your Data Modeling Work Through Open-Source SoftwareBlancas, Michael · PyData NYC 2022
• 38mJumping into Jupyter NotebooksJoey deVilla · DevFest Florida 2019
• 44mUsing Python with Hadoop to create production ready big data applicationsUlrich Zink · PyData London 2016