Sanyam Bhutani
149 episodes listed below
PRESS PLAY
Chai Time Data Science show is a series where Sanyam Bhutani interviews his Data Science Heroes: Practitioners, Kagglers & Researchers about all things Data Science
Based on 6 episodes we listened to · September 2026
Chai Time Data Science turns competition results and machine-learning careers into practical accounts of how models, experiments, and technical judgment develop.
Sanyam Bhutani keeps returning to the decisions beneath machine-learning results: how practitioners build a baseline, choose validation methods, test augmentations, assemble models, and move from competition experience to applied work.
Six recommender models enter; your neat single-model plan starts looking a little lonely.
Press play if you want
Short reviews from the PlayNext desk, based on the episodes we processed.
A practical panel of Kaggle legends comparing the model-training choices they actually use.
The panel brings together a set of Kaggle legends to weigh practical best practices, with the conversation centered on early stopping, learning-rate schedulers, data augmentation, and stacking.
A grounded Kaggle conversation about starting simple, moving fast, and paying attention to small validation choices.
Chai Time Data Science keeps the focus on process, not polish. The host starts with the return of audio interviews, the plan to bring live-stream conversations over from YouTube, and why those talks will still live on the podcast.
A steady conversation about podcasting as research and keeping a clear eye on AI hype.
Kathleen Walch and Ronald Schmelzer present AI Today Podcast as a long-running habit of asking vendors what they are actually building, and that framing gives the episode real shape.
A grounded look at how recommender-system wins are built under competition pressure.
This episode follows a recommender-system team through a hard, very specific deep-learning problem, and it stays grounded while doing it. It moves between GPU acceleration, multi-model stacks, feature engineering, embedding-matrix tricks, and ensemble design without drifting…
A career pivot opens into a practical tour of graph data structures and the Neo4j ecosystem.
Chai Time Data Science opens its second season with Dr. Claire Sullivan, a graph data science advocate at Neo4j, framing her move from nuclear engineering and academia into industry as the way in.
Emil Wallner frames internet-taught machine learning as a slow, self-directed practice shaped by motivation, rejection, and steady work.
Emil Wallner's path into machine learning comes through shifting interests, travel, and the discipline of learning online. Sayyam Bhutani lets that story unfold at an unhurried pace, linking Wallner's artist-in-residence work at Google with the projects and machine-learning…
A grounded look at moving from arts and languages into data science through steady Kaggle notebook practice.
This single-episode review frames a career change into data science as the heart of the conversation. The guest's path from arts and languages to Kaggle notebooks gives the episode a clear throughline: curiosity, six months of focused learning, visible practice, and the slow…
A grounded look at the patience, debugging, and visualization behind computer vision research.
Laura Leal-Tache's visit to Chai Time Data Science gives the episode a clean throughline: how an interest in computer vision becomes a working research life.
AI research, explained by the people doing it.
Talking shop with the people behind data products
Specialists walking through software's tradeoffs, one system at a time.
Founders and engineers on AI coding agents like Claude Code, and supply chain security.
Jeremy Howard conducts a warm, wide-ranging interview with Sanyam Bhutani — freshly anointed Kaggle Discussions Grandmaster and community lead at Weights and Biases — tracing his path from a disenchanted CS student at SRM University in India to a respected voice in the global data science community.
Chai Time Data Science hosts a live panel on training models, and Sanyam Bhutani says the guests together hold 87 Kaggle gold medals.
Chai Time Data Science #159 opens with Sanyam saying he has not published an audio interview in six months and will move live-stream interviews into the feed starting next week.
["Kathleen Walsh and Ron Schmelzer of Cognolitica explain how AI Today Podcast has stayed weekly for four years and passed 200 episodes.", "Kathleen traces a path from marketing at Hard Hanks and large Bed Bath & Beyond and Buy Buy Baby accounts into AI-focused research, while Ron moves from MIT and Rodney Brooks to a dot-com startup that raised $60 million.", "They argue that AI projects fail when teams treat them like standard application development, ignore data quality and labeling, and skip methodologies such as CPMAI.", "They also describe podcasting as research: a way to cut through marketing spin, reach people who will not answer email, and collect real use cases from banking, construction, agriculture, and Amazon re:MARS."]
NVIDIA's ACM REXIS team says it won three recommender-system competitions in five months by combining XGBoost, PyTorch, TensorFlow, and two-stage stacking around Twitter's huge dataset.
Claire Sullivan of Neo4j recounts how an unnamed Nobel laureate’s insult about engineers influenced her decision to study nuclear engineering.
Sayyam Bhutani interviews Emil Wallner about self-teaching, machine learning, art, travel, and creating educational content online.
Sayyam Bhutani interviews Andrada Altiano about moving from statistics and a two-and-a-half-year data analyst role at Avon to data science and Kaggle Notebooks Master status.
Sayem Bhutani interviews Laura Leal-Taché, professor at the Technical University of Munich and head of the Dynamic Vision and Learning Group, about computer vision research and academic careers.
Yifan Sheh recounts moving from Airbus research and a 2017 Kaggle start to data-science consulting after becoming a Kaggle Master.
Robert Bracco describes leaving Andrew Ng's 2014-era Coursera course unable to build beyond an Octave digit recognizer, then finding Fast.ai's build-first approach.
Tim Dettmers describes moving from philosophy, psychology, statistics, and neuroscience into deep learning research at the University of Washington.