Mapping and modernizing a production real-estate data platform
Audited a live AWS data platform end-to-end, reverse-engineered its ETL and search path, and delivered a tiered set of client reports — the same findings written three ways for three readers, from decision-maker to engineer.
- Mapped a production AWS platform: 19 EC2 instances, 5 RDS PostgreSQL clusters, 8 Lambdas, and S3 across legacy and modern accounts.
- Documented a strangler-fig migration from monolithic cron/EC2 ETL to Python microservices on ECS/Fargate + Airflow.
- Built a BigQuery → PostgreSQL fact-loading pipeline (partitioned queries, service-account auth) into the production event tables.
- Ran a Solr search-gap analysis — 34 concepts not reaching the index across 18 MLS feed sources — and traced it to code logic, not missing data.
- Unified 186 attributes from 2,962 raw feed rows into a canonical schema, with a field catalog and rollout options ranked by effort and risk.
- Shipped an MCP server exposing property search to LLMs, plus a natural-language search app on the same index.
AWS (EC2 · RDS · Lambda · ECS/Fargate · Airflow · S3) · PostgreSQL · Google BigQuery · Solr / Sunspot · Rails · Python · MCP