Creative Engineering / 001

Ideas you can
step inside.

Interactive experiments where photography, computer vision, production systems, and large-scale simulation become something you can see and explore.

Enter the first experiment

Experiment 001 / Visual criticism

Composition Studio

A vision model analyzes each street photograph for its strongest compositional techniques. The image-specific readings and coordinates are stored in PostgreSQL, then rendered as animated overlays that reveal the lines, layers, color, rhythm, and visual relationships within each frame.

Guided analysis
01 / 12
Selected street photograph
Reading frame…

Experiment 002 / Load testing

Simulated City

I helped engineer and create SimulatedRides at Lyft to load-test its two-sided marketplace at massive scale. This interactive homage sends probabilistic riders and drivers through ramp, soak, and recovery. Read the Lyft Engineering article I co-authored

ReadyTest phase
Awaiting riderActive driver0 vehicles
© OpenStreetMap © CARTO
SIMULATION / PEAK_EVENT00:00
Requests / min0
Active trips0
P95 match
Success rate100%
Cancellations00.0% of requests

Downstream service telemetry

Synthetic requests generated by client actions
Total throughput0 RPS
Users & identity 0 Nominal
Pricing 0 Nominal
Rides 0 Nominal
Dispatch / matching 0 Nominal
Cancels 0 Nominal
Payments 0 Nominal
Notifications 0 Nominal
Resource Manager 0 Nominal
Modeled service fan-out based on client actions. Cancels receives traffic only when a rider cancels; Resource Manager receives traffic only as rider and driver resources are allocated or released.
01

Probabilistic clients

Riders request and cancel while drivers accept and complete trips according to configurable behaviors.

02

Ramp, soak, recover

The scenario gradually injects demand, holds its target, then drains synthetic clients safely.

03

Observable outcomes

Marketplace health emerges through throughput, match latency, active trips, and success rate.

Experiment 003 / Production systems

Site Control Room

A safe, live window into how this portfolio is built, tested, synchronized, delivered, and operated.

Since deploy
PUBLIC TELEMETRY / SECRETS REDACTED
ApplicationRails 8.1.3.1Ruby 4.0.6 · production
Releasev549Revision
Flickr catalog160 photosSynchronized
Test suite0 examples0 spec files · Not configured
Delivery topologyRequest path
Visitor
Cloudflare
CloudFront
Rails / Heroku
Static assets
CloudFront
Object storage
amazon / S3
Cache policy
1 year / immutable assets
Image pipelineCatalog delivery telemetry

A live snapshot of how Flickr metadata becomes a database-backed gallery. Web requests never call the Flickr API.

  1. 01Flickr API syncChecking freshness…Daily job
  2. 02PostgreSQL catalog160 records · 101 KB metadataAtomic
  3. 03Vision analysis cache160 current · 0 errors100%
  4. 04Rails gallery querySeeded random order · 20 records per pageDB only
  5. 05Browser image delivery160/160 photos have all four responsive sizeslive.staticflickr.com
Analysis coverage160 / 160
Rendition coverage160 / 160
PerformanceLive delivery telemetry
Browser navigationMeasuring…DOM timing
Page / databaseMeasuring…DB snapshot 2.7 ms
Transferred assetsMeasuring…Browser resource timing
Build inventory4.78 MB18 compiled files · JS 895 KB · CSS 239 KB
New Relic APMServer connectedChecking browser agent…
CircleCI insightsLoading build history…Build-duration history · credentials stay server-side
Source activityRecent GitHub changes
Contacting public GitHub API…
Delivery historyDeployments & build
Current releasev549

2026-08-31T20:22:25Z

Loading CircleCI deployment records…
Last browser refresh now

Have an interesting problem?

Let’s make something
people remember.

Start a conversation