Lightning Talks
summary
Multiple short projects report concrete results: a contribution to BigQuery DataFrames implements explode/unexplode for struct columns, compiles expression trees to Ibis then SQL and lands in the API; a dataframe browser Buckaroo shows 5,000-row scrolling, histograms, and a pluggable 99th‑quantile analysis and ships on pip; a 3D typeface pipeline uses Grasshopper + Rhino plus DALL·E depth estimation to make point clouds, shrinkwrap meshes and printable models; object‑annotation widget runs React inside Jupyter with synchronized Python state; supervised vs unsupervised color embeddings are evaluated at scale and humans sometimes spot the true color wheel; GPU+serverless benchmarks claim ~6× cost efficiency.
AI-written summary · about
chapters (15)
- 0:00BigQuery DataFrames intro
- 2:18Exploding struct columns
- 3:43BigQuery DataFrames internship
- 4:05Finance to 3D typography
- 6:532D images to 3D models
- 7:36Software versioning challenges
- 10:50Versioning alternatives
- 12:52Color perception in ML
- 15:42Supervised vs unsupervised color
- 17:30Buckaroo for data exploration
- 20:50Interactive AI annotation tools
- 23:25Jupyter notebook AI demo
- 26:26Big data GPU cost efficiency
- 29:15NumPy multiprocessing physics
- 31:19CUDA string handling issues
