Data & Labeling Founders

Bakouk is building Sifflet so data engineers and the people who use their dashboards see the same alert, lineage, and context when a number breaks.

Coe is trying to make sensitive production data safe for developers to use, so engineering teams can build and test software without waiting on access to the real thing.

Dilocker rewrote Weaviate's prototype in Go and bet on the HNSW algorithm while it was still a niche paper, so the open-source vector database was ready when ChatGPT and RAG arrived. As CTO he is also an outspoken user of AI coding agents.

Eskildsen built turbopuffer after doing the math on what vector search cost, and set out to make search over every byte affordable.

Handy turned an analytics-consultancy workflow into dbt, the default way many teams transform warehouse data as code.

Hansen is pushing multimodal AI teams to treat data curation and annotation as core infrastructure, not a spreadsheet side job.

Huber spent years taking machine learning from demo to production at Standard Cyborg, and he now runs Chroma, the open-source database many developers use to give AI apps search over their own data.

Kannappan is turning Patronus AI's research on how AI agents fail into training tasks and simulated software environments that AI labs use to train agents.

Kirwan turned Uber-scale experimentation pain into Bigeye, a data observability and AI trust platform for enterprises that cannot afford silent pipeline failures.

Liberty built Pinecone so long-term memory for AI is infrastructure product teams can buy, not rebuild as research.

van Luijt built Weaviate so search over meaning stays open infrastructure teams can run themselves.

Masschelein is building Soda to catch bad data where it is produced, with checks and data contracts written by the engineers who own the pipelines.

Petrosyan is building SuperAnnotate around the expert human judgment that AI labs and enterprises need to label data and grade model answers.

Ramdas is trying to shrink the years between a promising semiconductor material and something a fab will put on a line.

Raymond is focused on the unglamorous bottleneck before retrieval: turning messy enterprise files into model-ready data.

Rotem spent years migrating huge telecom billing systems before co-founding K2view, and his idea that company data should be organized around each customer or other business entity runs through its Data Product Platform.

Segal wants data teams to run ingestion, reverse ETL, observability and cataloging in one system, so people asking AI tools questions can trust the data underneath.

Sharma built labeling software so computer-vision teams can train models without reinventing annotation infrastructure every time.

One of the original pandas authors is now building the data layer for images, video, and embeddings that tabular tools were never designed for.

Shmukler is bringing product-and-growth instincts from LinkedIn, Wealthfront, and Instacart to Anomalo's bet that AI can catch unknown data bugs before dashboards lie.

Tricot wants data integration pipelines to become a commodity teams can extend, instead of a one-off project for every SaaS source.

Vasnetsov built Qdrant because existing vector libraries were not enough for production similarity search at scale.

Wang built Scale around the idea that frontier AI is gated by high-quality data and evaluation, not just bigger models.

Xie built Zilliz and open-sourced Milvus to make vector search a first-class database category for AI applications.

Zayarni started Qdrant in 2021 with Andrey Vasnetsov after existing vector libraries could not power an HR matching product they were building. He keeps the open-source engine free of paid features and sells a managed cloud instead.