Skip to content

// THE TEAM

Who builds this.

We're not a 20-person firm pretending to be enterprise. PurviewX is one embedded operator who writes the code himself, using AI tooling he checks rather than trusts.

Alexander Snyder, founder of PurviewX

Alexander Snyder, MBA, CFI

Founder · Head of AI and Data Strategy

Embedded AI leader across six industries: energy, insurance, distribution, legal, enterprise security, and workplace safety. Systems running in production on three clouds, plus two products of his own. Also the author of the list of things that did not work, which is published here on purpose. Holds an MBA and a Certified Forensic Interviewer credential, which is where the habit of assuming a confident answer is wrong until it survives questioning comes from.

The work starts with joining the team: learning every name, attending every standup, understanding the business before touching the data. This isn't a philosophy. It's the only approach that consistently ships production systems.

Based in Portland, OR. Available for embedded engagements nationwide.

linkedin.com/in/alexandersnyder

// HOW THE WORK GETS DONE

The kinds of automation we build into client systems. These are patterns, not a product: each one is rebuilt for the client's own systems, and there is no shared agent platform behind them.

Data Pipeline Agents

ETL, enrichment, and deduplication, built to cache results before spending money, monitor yield per batch, and stop when yield drops instead of when errors appear.

Verification Agents

Checks that compare a system's output against its source and flag what does not match, before it reaches a report or a decision.

Analysis Agents

Pattern and anomaly detection over large operational datasets, aimed at the contradictions between systems that are supposed to agree.

Integration Agents

Connectors that reconcile several platforms into one operational picture. The largest so far unified five.

Ed

Edwin

Pre-release product

Edwin is a local-first AI memory product that Alexander Snyder is building. It stores each fact with the date it was true, keeps the reasoning behind decisions, checks new facts against what it already holds, and exports to AI tools over the Model Context Protocol. It is not yet generally available, and it has its own site at heyedwin.io.

Why it is on this page: Edwin is held to the same rule as client work. Its benchmark is pre-registered and self-run, and the harness, judge prompts and cost ledger are published. On answer accuracy, that benchmark found Edwin statistically tied with simply handing the model the full history. Its case is cost, not accuracy, and heyedwin.io says so.

A provisional patent application covering Edwin was filed with the USPTO in February 2026 (application 63/991,739), with 10 claims validated against 35 experiments. Two of those experiments disproved what we expected, and the filing says so. A provisional application is not a granted patent, and we are not going to let it read like one.

heyedwin.io

35

patent experiments

2

of those disproved us