NYC Taxi Fare Prediction on GCP

NYC Taxi Fare Prediction: GCP MLOps Pipeline

A cloud-native pipeline that goes from BigQuery public data to a deployed model. Features are built with SQL and dbt, an XGBoost model is trained, and a FastAPI backend and Streamlit frontend both run on Cloud Run. All infrastructure is defined in Terraform.

Repo: github.com/aksh-ay06/Trip-Cost-Prediction · Stack: BigQuery, GCS, dbt, XGBoost, FastAPI, Streamlit, Cloud Run, Kestra, Terraform, Docker

---

Pipeline

flowchart TD
    BQ[BigQuery public data] --> GCS[GCS parquet]
    GCS --> RAW[(taxi_raw)]
    RAW --> STG[(taxi_staging)]
    STG --> DBT[dbt: stg → fct_taxi_features]
    DBT --> TR[XGBoost training]
    TR --> ART[GCS model artifact]
    ART --> API[FastAPI on Cloud Run]
    UI[Streamlit on Cloud Run] --> API

Highlights

• Infrastructure as code: Terraform creates the buckets, BigQuery datasets, Artifact Registry, Cloud Run services, and IAM bindings.
• Layered warehouse: data moves through raw, staging, and production layers, and dbt builds the staging views and feature mart.
• Orchestration: Kestra runs each pipeline step in its own Docker container.

---

← All Projects