Jinja + YAML Manifests
Analysts describe cohorts declaratively using familiar Jinja constructs, enums, and loops. The CLI renders manifests with deterministic hashes for promotion and audit.
Streaming Extraction
SQLAlchemy-backed JDBC connectors stream warehouse data directly to Parquet without intermediate temp files, respecting chunk sizes and schema validation.
Flexible Scoring
Run scoring locally via Docker or integrate with a remote Node service that accepts payload uploads and streams back risk outputs.
Key Capabilities
- Manifest Rendering: `hcc-cli render` converts Jinja/YAML configs into normalized JSON/YAML manifests with SHA-256 digests.
- Extraction Pipeline: Run manifests through JDBC streaming to produce Parquet artifacts per template (members, diagnoses, etc.).
- Scoring Orchestration: Configurable Docker and HTTP clients submit payloads, capture outputs, and record metadata.
- CLI UX: Commands for dry-run previews, manifest emission, and full execution with environment variable injection.
Snowflake-Native & Ops Ready
The CLI mirrors Launch scoring UX while leaning into Snowflake Container Services best practices so analysts can move between portal and CLI without losing context.
- Hardened containers: Non-root execution, read-only layers, and enforced resource limits ship by default.
- Health instrumentation: `GET /health`, `SCORE_SERVICE_HEALTH()`, and `SYSTEM$GET_SERVICE_STATUS()` keep ops teams informed.
- Last run visibility: `GET /info` and Snowflake `SCORE_LOG` snapshots expose the most recent execution for dashboards.