Skip to content
Chew Shen
Resume
All work

2026In progress

FoodHunt v2

A rebuild of FoodHunt as a repeatable OpenStreetMap data pipeline and API.

Python, Django, GeoDjango, Django REST Framework, PostgreSQL, PostGIS, Docker, GitHub Actions

Problem
v1's OpenStreetMap import was a one-off script with no history, no data checks and no safe way to re-run it.
Decision
Make the pipeline the product: repeatable runs, raw data kept, and spatial queries in PostGIS.
Result
Foundation complete with CI; the pipeline, API and deployment are next.

The problem

v1's OpenStreetMap import was a one-off script with no idempotency, no history and no data checks.

What I built

  • Planned pipeline: keep each raw extract per run, transform it with pure tested functions, upsert on the OpenStreetMap id, and record every run.
  • Done so far: GeoDjango on PostgreSQL and PostGIS in Docker Compose, configuration from environment variables, a health endpoint, and CI with tests, linting and a migration check.
  • Every significant choice is recorded in a decision log.

Key decisions

  • Make the pipeline the product

    Restaurant finders are a common portfolio project, and the interesting engineering in v1 was the import. So v2 is built around the ETL pipeline and the API that serves it.

    Trade-off: The user interface gets minimal attention.

  • PostgreSQL with PostGIS

    v1 stored coordinates as plain numbers, which rules out efficient "within X metres" queries. PostGIS adds real spatial indexes and distance queries.

    Trade-off: The official image has no Apple Silicon build, so it runs slower under emulation locally, and hosting is limited to providers that support PostGIS.

  • Keep raw data and upsert on the OpenStreetMap id(planned)

    v1 matched places by name and area, so two branches of a chain in one area merged into one, and changes in OpenStreetMap never reached the database.

    Trade-off: Runs become repeatable and replayable, but raw storage grows with each run and will need a retention policy.

Results

  • Foundation complete. Next: the pipeline, data checks, scheduled runs, a geo-query API and deployment.