Upriver raises $14M to fix the unglamorous layer where enterprise AI quietly breaks: the data

Most enterprise AI projects do not fail because the model is bad. They fail because the data feeding it is a mess: broken pipelines, mismatched systems, and context locked in one engineer’s head.

Upriver, an Israeli startup, has raised $14M to automate the cleanup, betting that this dull but critical layer is where the AI era is really won or lost.

The seed round was led by Valley Capital Partners and Hetz Ventures.

Just as telling is the angel list, which reads like a roll call of the data-tooling world: Lew Cirne, who founded the observability giant New Relic; Abe Gong of the data-quality project Great Expectations; and the founders of the Israeli data-security unicorn Cyera.

Upriver says it is already used by Unity and the media group DMGT, and partners with Databricks and Snowflake.

Plumbing for the AI era

Upriver pitches itself as an “AI data engineering platform”: an agent that connects to a company’s full data stack, Snowflake, Databricks, BigQuery, Airflow, dbt, then explores it, builds and validates pipelines, fixes them when they break, and encodes the “tribal knowledge” that usually lives in people’s heads.

The promise is that data engineers stop spending their days investigating broken pipelines and stitching together tools that were never built to talk to each other, and instead decide what the data actually means.

The founders come at it from an unusual place. Ido Bronstein, the chief executive, and Omri Lifshitz, the chief technology officer, spent a decade building large-scale intelligence systems, work Business Insider reports was done for the Israeli military, before concluding that every company on a modern cloud stack was living the same problem they were.

The new funding will go into engineering, sales, and enterprise deployments.

The timing fits a broader correction. After two years of spending on models and chips, companies are scrutinising what AI actually returns, and a recurring answer is that it falls over on bad data. A wave of startups, from Capsa in private equity to Upriver in the data stack itself, is selling the same underlying promise: clean, trustworthy data is the thing standing between an AI pilot and something that works.

It is a $14M seed, early and unproven at scale. But the bet, that the foundation matters more than the model, is one a lot of money is starting to share.