Shaped

Shaped operates a real-time retrieval and ranking platform that combines vector search with behavioral feedback, positioning itself as a vector database with a feedback loop. Customers connect existing data sources and Shaped handles embeddings, model training and data freshness, returning ranked results in milliseconds. Queries can be issued by text, user ID or item ID, and the platform is used for personalized "for you" feeds, hybrid keyword-plus-semantic search, similar-item recommendations, personalized email content, AI assistants and contextual memory retrieval for agents.

The product is organized in three layers: a data layer with more than 30 native batch and streaming connectors (Snowflake, BigQuery, Redshift, Databricks, Amplitude, Segment, Rudderstack, PostHog, Kinesis, Kafka, Pub/Sub, Postgres, MongoDB, Shopify, and per earlier coverage MySQL, Google Analytics and Mixpanel) plus unified schemas and SQL views; an intelligence layer handling dynamic embeddings, continuous training and user/item context models; and a query layer exposing ShapedQL, a SQL-dialect interface that compiles to a multi-stage retrieve-filter-score-reorder ranking pipeline. Access is also available through Python and TypeScript SDKs and an MCP interface. The company describes the platform as developer-first, letting teams choose data sources, integration methods, language models (such as Llama, CLIP and BERT) and scoring mechanisms, with a dashboard for testing models and inspecting why results are recommended.

Shaped states SOC 2 Type II certification, GDPR, CCPA and HIPAA compliance, a 99.95% uptime SLA, sub-50ms latency and support for 1,000+ QPS.

Founding story

Shaped was founded in 2021 by CEO Tullie Murrell, CPO Daniel Camilleri and Shai Bruhis, who has since moved on to another venture. Murrell previously worked at Meta/Facebook AI Research, where he created PyTorchVideo and was an original core contributor to PyTorch Lightning; Camilleri previously worked at Uber and Afterpay. Murrell and Camilleri first met about a decade earlier at a student hackathon in Australia. The stated founding aim was to let small businesses build hyper-personalized experiences comparable to big tech without employing large machine learning teams.

Business model

Shaped sells its relevance platform to businesses as a hosted developer service. Pricing is described as modular, with a fixed platform fee plus tiered usage fees, and unifying search and recommendations on the platform is said to carry no extra cost. A self-serve entry point offers $100 in free credits without a credit card.

Subscription/usage-based software: a fixed platform fee combined with tiered usage-based fees, alongside a self-serve tier with free starting credits.

Traction

The site cites 1B+ documents and 100M+ users, enterprise scale of 5B+ documents, and customer case-study outcomes including 2.5x engagement, +79% D6 retention, +131% and +10% watch time, +16% AOV, 2.2x conversion, +9% purchases, +8% CTR and +382% diversity. It also claims a 10x increase in experimentation velocity and roughly seven days from data connection to a first production experiment. Y Combinator lists a team size of 25.

Latest developments

Shaped launched ShapedQL, a SQL dialect purpose-built for agents, feeds and search that compiles to a multi-stage ranking pipeline, together with Python and TypeScript SDKs, MCP support and a playground demo. The company's website announces that Whatnot has acquired Shaped, pointing to TechCrunch coverage of the transaction.