
Aaryan Gupta
Senior Software Engineer

Antonio Momblan
Software Engineer

Chirag Agrawal
Senior ML Engineer


Francisco Arceo
Senior Principal Software Engineer

Jim Dowling
CEO & Co-Founder

Mikael Ronström
Head of Data

Nikhil Kathole
Principal Software Engineer

Nilotpal Pramanik
Software Development Engineer III

Paarth Chothani
Staff Software Engineer


Pengyu Hou
Senior Software Engineer

- RM
Rishabh Mehrotra
Co-founder and CEO of Pavo AI

Roman Grebennikov
Principal ML Engineer


Tom Kaltofen
AI Engineer

Varant Zanoyan
Co-Founder
- VZ
Vitalii Zhebrakovskyi
Senior Software Engineer

Will Burstein
Head of Product


William Edwards
Staff Data Engineer


Yash Shukla
Software Development Engineer II

Yuli Han
Senior Software Engineer

- Oct 6, 03:30 PM UTC · 10 min · Opening
› Opening
Jim Dowling opens Feature Store Summit 2026.
- Oct 6, 03:40 PM UTC · 30 min · Presentation
› Beyond Feature Retrieval: Transformations and Composition in Feature Views
Why feature stores that stop at storing and retrieving features leave the model-ready feature vector on the table, and how Feature Views compose, join, and transform it.
- Oct 6, 04:10 PM UTC · 20 min · Presentation
› Feature Serving Beyond RAM
The long tail of online features sits in Redis at RAM prices. Murrdb, an open-source columnar cache, tiers hot to RAM, warm to NVMe, cold to S3.
- Oct 6, 04:30 PM UTC · 20 min · Presentation
› Taming the ML Firehose: Scaling Feature Consistency
Ranking and search models need fresh, consistent features. When production values drift from training, models degrade. How Uber keeps them in sync.
- Oct 6, 04:50 PM UTC · 20 min · Presentation
› Designing an Efficient Real-Time Feature Store at Zomato
Zomato's Profile Store redesigned to serve ~40M RPM with lower cost and latency: protobuf, DynamoDB and Redis tuning, and request coalescing.
- Oct 6, 05:10 PM UTC · 20 min · Presentation
› To be announced
Talk and speaker to be announced.
- Oct 6, 05:30 PM UTC · 20 min · Presentation
› OpenAI-Compatible Agents with Governed Context: Feast, MLflow, and OGX
Most agent stacks stop at retrieval. This one closes the loop end to end with Feast, MLflow, and OGX: served context, traces, labels, and a fine-tuning set.
- Oct 6, 06:00 PM UTC · 20 min · Presentation
› Accelerating and Protecting Payments with Real-Time Data for ML
Adyen's ML models drive fraud detection, routing, and payouts. How the Feature Platform generates real-time data at global scale inside the payment engine.
- Oct 6, 06:20 PM UTC · 20 min · Presentation
› Real-Time Features without Streaming using RonDB
Real-time feature computations natively in RonDB: rolling aggregations at request time in single-ms latency, no Flink pipeline to operate.
- Oct 6, 06:40 PM UTC · 20 min · Presentation
› Random Access Parquet: Serving Point Queries Straight from the Data Lake
Look up a single key directly in the Parquet files you already have. An external index turns a lookup into one small ranged read instead of a scan.
- Oct 6, 07:00 PM UTC · 20 min · Presentation
› Review Loops for Agents Using Real-Time Context
Agents are only as reliable as the context they retrieve. A practical review loop: traces, scorecards, human plus LLM-as-judge, and release decisions.
- Oct 6, 07:20 PM UTC · 20 min · Presentation
› Online, Offline, n-Chaos: From Stored Features to Executable Definitions
Feature stores keep online and offline consistent inside their boundary. What reliability looks like outside it, when agents walk into n versions of a feature.
- Oct 6, 07:50 PM UTC · 20 min · Presentation
› The Benchmark to Production Gap: Frontier Evaluations for Adaptive Personalization
Teams evaluate hundreds of ideas offline but deploy a handful, and conventional metrics pick losers. Frontier benchmarks lifted offline-online correlation from -0.4 to +0.8.
- Oct 6, 08:10 PM UTC · 20 min · Presentation
› Feast vs Chronon: Why Not Both?
Feast's developer experience and Chronon's production feature engine in one workflow, shown end to end with a running checkout-risk demo.
- Oct 6, 08:30 PM UTC · 20 min · Presentation
› From Daily Batch to Real Time: An NRT Generative Recommender with Chronon
Two new Chronon capabilities, Push Mode and NRT Model Transform, that turn a model's output into just another feature, kept continuously fresh.
- Oct 6, 08:50 PM UTC · 20 min · Presentation
› How to Know You've Outgrown Your Real-Time Feature Store
Feature stores rarely fail all at once. The warning signs of a structural platform problem, and a phased migration that reduces risk.
- Oct 6, 09:10 PM UTC · 20 min · Presentation
› Semantic IDs for Recsys with Chronon
A technical deep dive into semantic IDs in recommender systems: retrieval, ranking, cold-start, freshness, and avoiding training-serving skew with Chronon.
- Oct 6, 09:30 PM UTC · 20 min · Presentation
› To be announced
Talk and speaker to be announced.
- Oct 6, 09:50 PM UTC · 10 min · Wrap up
› Wrap-up
Jim Dowling closes Feature Store Summit 2026.