About Feldera
Practitioners building beyond stale features
Real-time ML needs real-time features, not batch approximations. Feldera bridges the delay between data arrival and model action.
Origin Story
Why Feldera exists
Lalith Suresh and Stefan Braun spent years building distributed stream-processing and database infrastructure. In several large-scale ML deployments, they saw models trained on hours-old or days-old data receive features from the previous night's batch job. The model knew this morning's fraud patterns. The feature store still showed yesterday's transaction totals.
The usual fix was adding orchestration around batch pipelines: stream a few critical features, maintain two codepaths, and tolerate consistency issues. None solved the root problem. The system was built for batch, with streaming attached afterward.
Feldera began with another premise: write SQL once, then let the engine update results incrementally, processing only changed rows as data arrives. No batch window, dual codepath, or designed-in staleness. That incremental SQL engine is Feldera's core, founded in 2023 and built through user adoption without outside funding.
Team
The people behind Feldera
Lalith Suresh
CEO
Years spent building distributed systems and stream-processing infrastructure across large-scale deployments. Focused on the gap between data availability and model freshness in production ML.
Stefan Braun
CTO
Deep experience with database internals, query compilation, and low-latency operator graph execution. Designed Feldera's core incremental SQL compilation pipeline.
Nadia Petrov
Head of Developer Relations
ML infrastructure practitioner turned developer advocate, with experience in open-source communities and technical education. Brings that perspective to Feldera's documentation and developer experience.
Values
What we believe
01
Correctness first
Incremental results must exactly match batch recomputation. "Close enough" does not work for feature data feeding a fraud model. Every operator is tested against a full-recompute baseline on every commit.
02
Developer experience matters
Adopt Feldera with SQL you already know, without learning a DSL or new query language. The engine handles incrementality, not the query author.
03
Built independently
There is no outside funding. We are building a sustainable product through user adoption, staying focused on engineers' needs rather than pitch decks or acquisition appeal.
Join us
Follow our build
We share what we build and why. Begin with the documentation, or read the engineering blog.