Eric Crook

Backend & AI engineer · San Francisco Bay Area

Backend systems.
Useful AI.
Work you can inspect.

I build the services and data pipelines behind products people use. My work spans production backends, analytics and AI tools that keep their evidence close.

Open to hands-on backend and AI engineering roles, including contract work.

Explore the workEmail Eric

Engineering work

Production experience and runnable examples, with the scope of each made explicit.

Production engineering

Raven Analytics

Rust · Polars · AWS · data correctness

A cheaper pipeline still has to produce the right answer.

I rebuilt a Python-on-EC2 analytics pipeline in serverless Rust and Polars. The measured workload ingested more than 2 GB of events a day, with roughly 90% less compute spend than the previous implementation.

For validation, I reconciled daily row counts exactly, checked column-level counts over matching date ranges, and investigated discrepancies against source logs. That work uncovered duplicate events in the original data.

These results describe that workload and comparison, not a guarantee for every analytics system. Raven is a product of my company, math.technology.

Explore a related correctness example

The synthetic daily brief below demonstrates the same principle: every number should have a definition, source and limitation. It is a new offline example, separate from the production migration.

Runnable synthetic example

A daily brief with evidence

Python · deterministic metrics · validation

Show what the data supports—and when it is incomplete.

This offline example computes net revenue and next-day retention from a small synthetic event export. It handles duplicate retries, rejects conflicting IDs and mixed tenants, and withholds retention rates for incomplete days.

Inspect contributing event IDs, source hashes, cost counters and the distinction between recommendations and actions. It calls no model or cloud service.

All users, events and dollar amounts in this demo are synthetic. They do not represent Sixth Sigil revenue or customer results. Retention uses the first session observed in the sample, not a verified signup date.

Open the synthetic daily briefInspect JSON evidenceReview the Python source and tests
Interactive UI prototypes

Interfaces for reviewing agent work

Two browser-only explorations of human review and debugging. Both use illustrative data; they are not connected to live agents or customer systems.

Greenlight approval queueInspect a proposed action, approve or reject it, then undo a decision.Stepwise trace inspectorExplore spans, latency and token usage in an illustrative agent run.

Independent products

Games and tabletop tools have their own space. Try the products directly.

Visit the games collection

Let’s talk about what you’re building.

I’m interested in useful products, difficult backend problems and teams that value shipping carefully.

Python, Rust, Go and TypeScript. APIs, event pipelines, AWS and practical AI systems.

erictcrook@gmail.comLinkedIn profileGitHub profile