TerraCore

I built a project called TerraCore, an API that takes a US address or coordinate and returns ~200 neighborhood signals — demographics, income, housing, walkability, nearby amenities, and other location intelligence — resolved at Census block-group precision (~1,500 people).

Earlier in my career I worked at CoreLogic, a large real estate data company, so I’d spent time involved with the creation and consumption of spatial data products. But this was the first time I built an entire geospatial data pipeline myself from scratch without corporate resources or budgets behind it.

Going into the project I assumed the hard part would be the geospatial work.

It wasn’t...

PostGIS actually makes the spatial side relatively easy compared to how things used to be done manually. Spatial containment, joins, proximity queries — those are well solved problems.

The real challenge was turning messy public datasets into something clean enough to build an API on top of.

Census Data Looks Simple Until You Use It

From the outside, Census ACS data looks structured and authoritative.

In reality it’s a massive set of tables with variable names like:

B01001_003E
 B19013_001E
 B25077_001E

Each table contains dozens or hundreds of fields, and the documentation often feels like reading a statistical manual written in ancient Egyptian hieroglyphics rather than something useful for developers.

Mapping those tables into something usable required going through hundreds of pages of documentation and rebuilding the data pipeline multiple times just to get the column mappings and derived metrics right.

Address Data Was Even Messier

For geocoding, one of my sources is OpenAddresses, which is an incredible public dataset, but the raw records can be chaotic.

-> Missing street suffixes
-> Malformed house numbers
-> ZIP codes in strange formats
-> Missing city, state, and zip all at once

At one point I had to rebuild and normalize hundreds of millions of address records, scrubbing tens of millions of malformed entries just on the second pass alone.

It was one of those moments where you realize the real work in data engineering isn’t the algorithms — it’s cleaning the data.

OpenStreetMap Has Its Own Challenges

OpenStreetMap is an amazing dataset, but its flexibility means the tagging system can be inconsistent.

The same real-world thing can be tagged multiple ways.

A grocery store might appear as:
shop=supermarket
shop=grocery
amenity=marketplace

And something tagged as a “deli” can mean completely different things depending on where you are in the country.

In New York most delis behave like sandwich shops. In other parts of the US they’re closer to small markets. That kind of ambiguity makes classification harder than it looks.

The big lesson

The biggest takeaway from building TerraCore was this:

Geospatial infrastructure isn’t really about spatial math — it’s about taming messy data.

The tooling around geospatial queries is mature. Spatial joins, distance calculations, containment queries — those are solved problems.

But turning chaotic public datasets into something a developer can drop into an application in an afternoon takes a surprising amount of unglamorous engineering work.

That’s essentially the problem TerraCore tries to solve.

Instead of wrestling with Census tables, address normalization, and OSM tag inconsistencies, the goal is to let developers just drop an address into an API and immediately get structured neighborhood context back.

What I’m still figuring out is where this kind of data is actually most useful.

My intuition is things like AI enrichment pipelines, real estate analytics, or location-aware apps that need fast context about a place. But that’s still a hypothesis.

So I’m curious:

If you're building something location-aware — real estate tools, logistics systems, local marketplaces, AI agents that need location data — what neighborhood or location data have you wished existed as a clean API? And how would you consume that data?

If anyone’s wants to look further into the dataset it’s called TerraCore and the API field list is here. I put it on RapidAPI as a soft start.