A few projects I’ve built on the Location Data team at Indeed. Where I’ve written a longer post, the title links to it.

For the longer story, see About me.

Carmen: places database editor

Carmen place detail view with an interactive map

A validated, transactional web app that replaced hand-written SQL for creating, editing, and deactivating point-of-interest records.

PythonFlaskMySQLLeaflet

AI location extraction pipeline

Daily pipeline: sample job postings, extract locations with an LLM, classify, match with a geocoder, and publish results for precision and recall tracking.

An LLM pipeline that replaced a manual vendor review task: ~$90K/yr saved, 120 → 4,200+ jobs/day, ~95% accuracy.

PythonGPTTF-IDFScheduled jobsAPIs

Geocoding credit & volume dashboard

Geocoding credit dashboard demo with sample data

Makes geocoding and autocomplete spend, cache behavior, and market usage visible, with a 30-day burn projection. Live demo (sample data).

Plotly DashPythonSQLBusiness Intelligence

Regression report toolkit

A proposed location database change is checked for hierarchy, duplicate, coordinate, and orphaned alias problems.

Self-service regression testing for location database changes, with SQL validation checks that block unsafe edits before they ship.

PythonFlaskMySQLDockerBashJava

Location labeling studio

Location labeling studio queue: the model's extracted address next to the job description

Human review of model-extracted street addresses: it allows for a shared queue with handling for concurrency, labeling, immutable revisions, and snapshots that feed model training.

PythonFlaskTrino

Transit updater: Japan transit data

A quarterly EKI snapshot is diffed, screened by anomaly detectors, turned into Flyway SQL, reviewed by an analyst, and validated.

Turns each quarterly Japan transit data (EKI) refresh into reviewed, validated SQL migrations with zero service disruptions.

PythonPostgreSQLFlywaypytest