Shaped.ai

Interface & Interaction Design | Prototyping | Front-End | Consulting | Strategy

Shaped is a vector database and self-serve platform that provides API's to power AI-driven recommendations and search. This is most critical to products and platforms that are marketplaces, social media, media consumption, or e-commerce. I was brought on by the founders to support them as their product design capacity. Our work together is ongoing and has primarily focused on re-architecting and redesigning their console (formerly dashboard) to ensure it scales, establish their core experiences, providing cohesive design strategy and vision, and enable them to move faster by helping build the front-end implmentation.

Together with the team we dove into customer research to understand needs for both self-serve and white glove customers. Technical customers can use the CLI or SDK, while less-technical customers can use the console to setup, evaluate, and deploy their own recommendation and search models. In the console, engines can be queried to provide a clear sense of relevance, rank score, comparative model performance and more.

Onboarding

  • Shaped sign-up page with GitHub, Google, email, and AWS marketplace options
  • Email verification step during account creation
  • Home dashboard with getting started guides, API keys, team invites, and resources

DAta

Users can view and manage ingested tables and views (aka data transforms). The tables experience guides users from first connecting a data source through to inspecting schema, column-level statistics, health, and raw records. Users can upload local files, connect to sources like BigQuery, Snowflake, or Segment, or get started with sample datasets.

  • Tables empty state with data onboarding options
  • Manual table creation with namespace and field mapping
  • Data connector picker with BigQuery, Snowflake, Segment, and more
  • Connector setup flow with field mapping and credential configuration
  • Populated tables list with status, health sparklines, and record counts
  • Table detail view with schema, column statistics, and raw data

Engines

Engines are the recommendation and search models that power Shaped's API. Users create engines from their ingested tables, configure vector and lexical search, and select embedding models. Once deployed, engines can be monitored through activity charts segmented by user demographics and offline metrics that compare model performance against random and popularity baselines.

  • Engines empty state with creation options
  • Engine creation with source table, search configuration, and embedding model selection
  • Engine cards with status badges, request sparklines, and context menu

Query & Results

With their engine created, users write, run, and evaluate their recommendation and search queries. Results can be viewed in multiple layouts including rich tables with column-level statistics, masonry product grids, and more. Users can inspect result explainability traces to understand how items were retrieved and ranked, and access API details for integration into their applications.

  • Simulated query results in masonry product grid layout
  • Rich table results with movie poster thumbnails and metadata
  • Expanded rich table with column-level cardinality, distributions, and derived columns
  • Results trace showing retrieval journey, scores, and rank breakdown
  • Inline query documentation with JavaScript and Python code examples
  • Edit query template dialog for customizing request parameters and format Request details panel showing the API endpoint, key, and cURL snippet

Monitoring Usage and Performance

  • Engine activity chart segmented by user demographics with session and popularity overlays
  • Offline metrics comparing MAP, NDCG, Precision, Recall, and Coverage across baselines
  • Query insights panel with per-request latency and usage breakdowns by engine
  • Data ingestion monitor showing event counts, record health, and sync status
  • Engine run health view with deployment status and error diagnostics

v1 Dashboard

  • Shaped Home View
  • Simulating a Session Query
  • Querying a model by user id
  • Querying a model by item id Configuring query results
  • Item metadata helpers User metadata helpers

Shaped Data Catalogue

The Shaped Data Catalog is a glanceable and powerful view, allowing users to see the data schema and leveraged datasets in their models and a clear view or the raw data, their top level metrics and health. Shaped has three schema types that are core to how people query. They are users, items, and events. Users will query the API or dashboard by these types when evaluating their model and the relevancy of search or recommendation results. The data catalogue also allows users to also monitor the health and integrity of their models and what data has been ingested and mapped. This is critical to understanding when deployments fail, why a model's performance may suffer, interpreting key metrics more clearly, and more.

  • Viewing a model's origin datasets and features
  • Seeing model runtime activity
  • Configuring query results by attributes Configuring complementary item results