TWIMLcon: AI Platforms 2021
| title | author | length |
|---|---|---|
| AI Operationalization: Where the AI Rubber Hits the Road for the Enterprise | Rasool Tahmasbi, Sarah Cullem, Conor Jensen | 29m |
Non-digital-native enterprises report concrete progress and remaining friction: Clorox runs one model in production (a marketing forecasting model) while Palo… | ||
| Algorithmia in Action: Take your ML models from training to production | Kristopher Overholt | 59m |
Algorithmia is an MLOps platform that demo deploys regression, gradient-boosting, image-classification, sentiment and time-series models from notebooks into a… | ||
| Build an End-to-End ML Workflow for Your Organization | Mohamed Elgendy | 28m |
Organizations need a centralized data lake, feature store and model registry, plus experiment tracking and GPU/distributed training to get reproducibility and… | ||
| Build, Buy and the Golden Ratio: How Theator scaled up its continuous training framework | Dotan Asselmann, Ariel Biller | 28m |
Company scales from 80 videos to 15,000+ procedures, ingesting hundreds more daily while automating nightly retrains and closed‑loop feedback. Uses ClearML for… | ||
| Building ML Platform Capabilities at Global Scale | Paul van der Boor | 29m |
Prosus teams build distinct platform pieces that actually run at scale: OLX pairs a searchable data lake and Airflow-driven reservoirs with KFServing for… | ||
| Building Teams and Culture that Support ML Innovation | Ziad Asghar, Pardis Noorzad, Ameen Kazarouni | 30m |
Argues that budget and clear metric hierarchies matter more than vague mission statements, and that success is measured by product impact — examples: reduced… | ||
| Building the Business case for ML Platforms | Kirk Borne, Justin Norman, Divya jain | 31m |
Argues that an ML platform proves value by concrete outcomes not promises: reduced deployment time 6–9 months to ~3 months and fewer duplicated models across… | ||
| ClearML: Your R&D on MLOps! From zero to hero in two lines of code | Ariel Biller | 53m |
ClearML is an open-source experiment + orchestration platform that the demo integrates with two lines of code to auto‑log runs, convert notebooks to runnable… | ||
| Cost Transparency and Model ROI | Srivathsan Canchi, Ian Sebanja | 30m |
Intuit runs 400+ models with 8,000 features on AWS using SageMaker, Kubernetes and Argo, and implements per-model asset IDs, fine‑grained tagging and BU… | ||
| Data Prep Ops: The Missing Piece of the MLOps Puzzle | Jennifer Prendki | 30m |
Claim: poor AI ROI comes from failing to treat dataset work as engineering. Company Alectio builds a pipeline that prioritizes data selection (useful vs… | ||
| Deploying a Fraud Detection Model with a Feature Store | David Hershey | 01h 52m |
Builds a real-time fraud pipeline end to end using a Tecton feature store on synthetic PSIM transactions: defines features (credit_score, age_in_days… | ||
| End-to-End ML | Jeff Fletcher | 53m |
Cloudera shows an end-to-end churn pipeline that claims only 20% of models reach production, built on Cloudera Data Platform and CML using Kubernetes… | ||
| Executive Summit Keynote: How to Ensure ML Project Success | Franziska Bell | 35m |
Executives get concrete actions: start with business pain not algorithms, pick KPIs that measure end‑user actionability (avoiding the common 60% failure rate… | ||
| Executive Summit Roundtable Discussion | Ameen Kazerouni, Hussain Mehanna, Paul van der Boor | 33m |
Building an ML org in a greenfield at Orangetheory means instituting a shared language, peer-reviewed dataset design, a feature store and model registry, and… | ||
| Explainable Monitoring for Successful AI Deployments | Krishna Gade, Rob Harrell | 45m |
A monitoring system called Fiddler injects explainability into production models by computing drift with Jensen-Shannon divergence and attributing it via a… | ||
| Hands-On Feature Store and Model Deployment | Monte Zweben, Jack Ploshnick | 02h 33m |
Splice Machine runs a single hybrid DB (OLTP + OLAP) on Kubernetes with Spark and an mlflow fork to build a feature store and deploy models as in‑database… | ||
| Harnessing On-Device AI to Transform the Smartphone Experience | Vidyasagar Rao | 54m |
Qualcomm shows on-device pipelines that move camera, audio and NLP from cloud to phone: a portrait-relighting flow combining demosaic → denoise →… | ||
| How Feature Stores Enable Operational ML | Mike Del Balso | 30m |
Data teams face duplicated pipelines and stalled models; feature stores automate feature transforms, store both online and offline feature values for training… | ||
| How Spotify Does ML At Scale | Aman Khan, Josh Baer | 29m |
Spotify reports building an ML platform since 2017 that centralizes tooling—Jukebox feature store, Kubeflow Pipeline (skf/qflow), Salem model serving and ML… | ||
| Keynote Interview | Chris Albon | 56m |
Wikimedia runs about 100 ORES models for edit quality, topic and damage detection and serves them at Wikipedia scale from on‑prem hardware rather than public… | ||
| Keynote Interview | Faisal Siddiqi | 53m |
Netflix runs a loosely coupled, seven‑layer personalization platform supporting model development, feature engineering and Java microservice serving for… | ||
| Keynote Interview | Solmaz Shahalizadeh | 51m |
Describes how Shopify moves from one model to hundreds by embedding data scientists in product teams, building an in‑house Python and Spark platform since… | ||
| Keynote Interview | Ya Xu | 52m |
LinkedIn runs about 500 concurrent experiments to measure real-world impact, maintains a central data science org of ~300–350 people that embeds with product… | ||
| Machine Learning is Going Real-Time | Chip Huyen | 25m |
Real-time ML splits into Level 1: online predictions (ms–1s latency) and Level 2: online learning (minutes, micro-batches of ~100–500); TikTok uses real-time… | ||
| Making Models Work: How an optimized ML lifecycle can drive better ROI | Priyank Patel | 32m |
Shows a platform play with concrete results: a bank uses 2 PB of data to cut AML false positives up to 60% and then scales the same pipelines to 25 teams for a… | ||
| ML Product Experiments at Scale | Justin Norman | 26m |
Yelp runs a homegrown experimentation platform called Bunsen that serves ML and product tests across production, running about 700 experiments concurrently and… | ||
| MLOps for High-Stakes Environments | Sudeep Pillai | 31m |
Toyota Research Institute builds an MLOps pipeline that turns vehicle logs into labeled scenarios, trains models, runs quantized TensorRT builds on… | ||
| One Flow to Rule Them All: Building the end to end ML project to production | John Posada | 01h 03m |
Demos building an end-to-end ML flow inside Dataiku DSS that connects 40+ data sources, mixes visual recipes and code notebooks, trains models with automl or… | ||
| Operationalizing Responsible AI | Sarah Bird, Kathy Baxter, Romer Rosales | 32m |
Teams describe concrete practices: a weekly ship room where product teams, fairness experts and OKRs force a documented fairness plan and systematic testing… | ||
| Team Teardown: Driving Platform Adoption and Success | Maya Hristakeva, James Kirk, Lex Beattie, Maisha Lopa, Samuel Ngahane | 30m |
Spotify describes a production ML stack built around kubeflow (SpotifyQFlow) plus a shared cluster and CLI, plus serving and feature products named Salem… | ||
| Team Teardown: MLOps in the Cloud for IoT | Danielle Dean, Mathew Salvaris, Mohan Mupiddi | 29m |
Built a Kubeflow Pipelines stack on Kubernetes with on‑prem GPUs, VS Code devcontainers and AWS Batch for synthetic jobs, supporting image models, RL… | ||
| Top Trends in Enterprise Machine Learning for 2021 | Diego Oppenheimer | 31m |
Algorithmia reports a third annual enterprise survey of mostly >$100M firms and claims a 76% increase in data scientists year over year, yet deployment gets… | ||
| Unified MLOps: Feature Stores & Model Deployment | Jack Ploshnick, Monte Zweben | 31m |
Splice Machine claims a unified HTAP SQL engine that combines feature store and model deployment so models run inside the database without endpoints or… | ||
| When Good Models Go Bad: The damage caused by wayward models and how to prevent it | Todd Underwood | 30m |
Hand analysis of ~100 incidents over 10 years shows many failures are infrastructure handovers not modeling math. A large ranking pipeline with thousands of… | ||
| Using Unity to Generate Synthetic data and Accelerate Computer Vision Training | Jonathan Hogins, James Fort | 58m |
Unity generates annotated training data with the Perception SDK / perception package, using domain randomization to vary lighting, backgrounds, occlusion and… | ||
| Why ML Projects Fail and How to Ensure Their Success | Adrian Cartier, Andy Minteer, Jürgen Weischenberger | 32m |
Recent Gartner estimates put ML project failures at 60-85% failure rate and the panel gives concrete mitigations: treat projects like an investment portfolio… | ||
| nothing matches | ||