Chalk Real-Time Feature Platform

Chalk is a data and AI infrastructure company with offices in San Francisco and New York, led by co-founders Marc Freed-Finnegan, Elliot Marx and Andy Moreland. It builds infrastructure for making machine learning and agent decisions at the moment they are needed, such as fraud checks, underwriting and recommendations.

The platform is organised into Context, Runtime and Learning. Context combines a feature store, a federated query engine and low-latency serving (the site cites 5 ms serving) that pull data from sources such as Postgres, Stripe and Snowflake without ETL. Runtime covers model serving, GPU scaling, sandboxes and a model gateway, and Learning covers evals and notebooks. Customers named on the site include Whatnot, Socure, Sunrun, MoneyLion and Medely.

• Feature store: defines and serves features in Python and SQL for training and online inference.
• Real-time serving: low-latency retrieval with the site citing 5 ms serving.
• Federated queries: reads from sources such as Postgres, Stripe and Snowflake without ETL.
• Model serving and gateway: runs models with GPU scaling and routes LLM calls through a model gateway in the customer's cloud.
• Evals and notebooks: tracing, agent evals and notebooks for iterating on models.
• Deployment options: self-hosted or Chalk-hosted with observability.

• A Vietnamese bank or consumer-finance company scoring loan applications and card transactions in real time.
• An e-commerce marketplace serving personalised recommendations and dynamic pricing from live behaviour data.
• A fintech or e-wallet running fraud and account-takeover checks at payment time.

• Features and context are computed at inference time, so decisions use fresh data rather than stale batch tables.
• Data sources connect without separate ETL jobs, which shortens the path from prototype to production.
• Can be self-hosted or Chalk-hosted, which suits teams with data residency needs.
• One platform for features, model serving and evals reduces the number of tools to operate.

Fraud detection, credit decisioning and personalisation increasingly depend on features computed in milliseconds, and AI agents add a need for live business context. Real-time feature platforms are becoming a standard layer in ML stacks. In Vietnam and Southeast Asia, banks, e-wallets and e-commerce operators are expanding online lending and risk scoring, which raises demand for this layer.

Offices in San Francisco and New York (per the About page). Raised a 10 million USD seed in December 2023 and a 50 million USD Series A in May 2025 led by Felicis. Named to Fast Company's Next Big Things in Tech 2025.