Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
75 lines (64 loc) · 2.3 KB

ARCHITECTURE.md

File metadata and controls

75 lines (64 loc) · 2.3 KB
Raw

Architecture

This project is a synthetic customer growth and pricing intelligence platform. It keeps the local path fully reproducible while documenting the cloud path a fintech team would use for a production version.

Local Product Architecture

flowchart LR
    generator["Synthetic event generator"]
    raw["Local parquet raw zone"]
    duckdb["DuckDB warehouse"]
    dbt["dbt staging, intermediate, and mart models"]
    dashboard["Streamlit product dashboard"]
    api["FastAPI prediction and scenario contracts"]
    scoring["Batch activation scoring"]
    pricing["Pricing scenario runs"]
    monitoring["Monitoring and calibration reports"]

    generator --> raw
    raw --> duckdb
    duckdb --> dbt
    dbt --> dashboard
    dbt --> api
    dbt --> scoring
    dbt --> pricing
    scoring --> monitoring
    pricing --> monitoring
    api --> monitoring
Loading

Cloud-Ready Target

flowchart LR
    source["Synthetic source events"]
    storage["Cloud Storage raw bucket"]
    bqraw["BigQuery raw datasets"]
    bqmarts["BigQuery dbt marts"]
    cloudrunapi["Cloud Run API service"]
    cloudrunjobs["Cloud Run batch jobs"]
    reports["Monitoring report artifacts"]
    streamlit["Streamlit dashboard"]

    source --> storage
    storage --> bqraw
    bqraw --> bqmarts
    bqmarts --> cloudrunapi
    bqmarts --> cloudrunjobs
    cloudrunjobs --> reports
    bqmarts --> streamlit
    reports --> streamlit
Loading

Product Surfaces

  • Streamlit dashboard: product health, pricing intelligence, experiments, and monitoring status.
  • FastAPI service: activation, churn, upsell, offer recommendation, and pricing scenario contracts.
  • Batch artifacts: activation score extracts, model monitoring reports, pricing scenario runs, and sensitivity CSVs.
  • dbt warehouse: trusted marts for activation, retention, engagement, pricing, experiments, finance, and geo incrementality.

Governance Boundaries

  • All data is synthetic.
  • The activation model is used for helpful onboarding prioritisation, not credit, eligibility, account limits, or punitive customer treatment.
  • Pricing outputs are synthetic offer and incentive scenarios, not regulated credit pricing or real personalised pricing.
  • Vulnerable-customer flags, complaint rates, support load, calibration, and drift are treated as release gates.