Database and data platform work for teams that already have a problem.
An independent practice focused on the unglamorous middle of the data stack, the migrations, the pipelines that have to run at 3am, and the production databases nobody wants to be paged for.
Services
Database engineering
Schema design, indexing and query tuning, and taking ownership of production instances that have outgrown whoever set them up. Across relational and document stores.
Migration
Moving databases between engines and between clouds, including the parts that are hard: the cutover plan, the rollback, and proving the data arrived intact.
Pipelines and ETL
Ingestion and reporting pipelines built to be re-run, monitored, and handed to someone else. Including replacing legacy jobs whose original author has left.
Platform automation
Infrastructure as code, CI/CD for database change, and identity/access wiring across cloud providers so environments stop drifting apart.
Capabilities
- Relational
- SQL Server, Azure SQL, PostgreSQL, Oracle, MySQL
- Non-relational
- MongoDB
- Cloud
- Azure, AWS, Google Cloud — including cross-cloud identity federation
- Automation
- Terraform, Ansible, GitHub Actions
- Reporting & ETL
- SSIS, SSRS, and the usual assortment of things that replaced them
- Applied AI
- Structured extraction and cataloguing pipelines, built for production rather than demo
Approach
Most data problems are not modelling problems. They are the accumulated result of decisions that made sense individually and stopped making sense together. The work usually starts by reading what is already there rather than proposing what should replace it.
Engagements are scoped to leave behind something your team can operate without the consultant who built it — documented, in version control, and boring to run.
Work
Three systems built and run end to end: schema, ingestion, spatial or physical modelling, tests, deploy, monitoring, backups. Each stays within a free tier, and each publishes what it is unsure about rather than hiding it.
Deedective — property intelligence for Greater New Haven, Connecticut. Parcel boundaries joined to FEMA flood zones, recorded sales, wetlands, septic suitability, water and sewer mains, a modelled airport-noise contour and a federal house-price index, answered as a per-address report that says where its data runs out. 18 towns, 225,000 parcels, 208,000 sales, PostGIS on one free-tier VM with nightly backups and a tested restore. deedective.com →
ChargeCliff — EV fast-charge time and highway range computed from each car's own charge curve rather than a brochure figure: a per-percent session integration, a cold-weather intake limit, drag scaling and a temperature-dependent heater term. 42 vehicles, four calibrated to named dyno benchmarks and the highway model pinned in CI to nine published 70 mph road tests; a highway-model error of 27% was found by that validation and fixed. Public API, rate-limited, least-privilege service. chargecliff.com →
PermitParser — an ETL platform that turns fractured municipal
zoning and permit portals into one queryable spatial warehouse. Scrapers that respect
robots.txt, parsers kept pure so every portal quirk is a fixture, Alembic
migrations, an arq worker, and tests against a real PostGIS in CI.
Live for eight towns (Branford, Madison, Middletown, Milford, North Branford,
Orange, West Haven, Woodbridge), 297 documents parsed into 190 filings, each geocoded and dated
to its meeting, eight parser conventions (three of them OCR of scanned agendas) behind one
record shape.
permitparser.com →
Underneath all three: OIDC federation to three clouds with no long-lived keys, budget alerts, a reproducible three-machine workstation bootstrap, and a daily health monitor that classifies its own failures. Total cloud spend across everything: about $25 a month, most of it one small VM and a handful of DNS zones; the largest dataset runs on a free-tier instance.
Contact
Available for scoped project work and advisory engagements. The most useful first message describes what is currently breaking, or what has to move and by when.
Connecticut, working remotely. Replies within a business day.