Brandon Bradley

I find the slow part.Then I fix it.

Staff software engineer with ten years in financial and analytics software. This year I also built and shipped products with AI, from the first prompt to the App Store.

Now
Moody's Analytics, since 2021
Before
The Vanguard Group, 2016 to 2021
Based in
Charlotte, NC, working remotely

The average client report took 30 minutes.

At Moody's, some ran so long they timed out. Those timeouts turned into incidents.

So I measured every step.

I rebuilt performance testing for the report library, from data fetching on AWS Lambda, to Redis, to writing the result to S3. That showed where the time went.

Then I optimized the Lambdas.

The longest report dropped to 5 minutes. The average fell to 3.

Incidents fell 75%.

Reports that finish don't time out. Faster reports meant more successful reports, and far fewer incidents to chase.

Ten years, to scale

Two companies, both in finance, with more responsibility at each step.

Vanguard5 years
Jul 2016 to Jul 2021
Moody's Analytics5 years and counting
Jul 2021 to now
Building with AIsee every build below
Mar 2026 to now

Moody's Analytics

Staff Software Engineer

Commercial real estate reporting

  • Cloud reporting platform. Led a Node.js, TypeScript, React and AWS platform that generates PDF and Excel reports for enterprise users.
  • Standards for AI coding. Wrote the rules for using Cursor and Copilot so AI-written code stays tested, readable and reusable.
  • Monitoring. Set up centralized Datadog monitoring and alerting.
  • Data. Built pipelines with PostgreSQL, DBT and SQL, and modernized large Postgres databases.
  • Accessibility. Led fixes across web and PDF reports that passed internal and external audits.
  • People. Mentored engineers and earned the Moody's Spotlight Award for coaching and support.

The Vanguard Group

Senior Software Engineer

Multiple promotions

  • Legacy to modern. Led an Angular application that replaced older systems, and directed several framework upgrades without breaking what worked.
  • Finance, made readable. Turned portfolio performance and financial analytics into D3 visuals.
  • Connected systems. Joined Java services to the front end through REST APIs.
  • Teaching. Served as the accessibility expert, trained teams, and helped new hires get up to speed.
  • Recognition. Vanguard Star award for commitment to excellence to crew and customers.

2026, built with AI

Every bar runs from first to latest work on that project. The amber mark is launch day. Select one to jump to it.

Take One, live since November 2, 2026

An AI editor that knows what to cut

Upload one rough take. Get back a finished short video.

  1. TranscribeWhisperX times every word.
  2. ReviewClaude finds the retakes, dead air and flubs to remove.
  3. DirectClaude plans which animation lands on which word.
  4. RenderOur own engine draws captions and graphics in the creator's brand.
  5. CheckA quality gate blocks the video if the pacing, timing or ending is off.
  6. DeliverThe creator gets the video and a caption for each platform.
2:11 → 0:28this clip, 20 cuts
about $0.40AI cost per video, tested
155animations built
Visit Take One

More that shipped

Luma, March to June 2026. Sold.

A meeting copilot I built and sold for $15,000

The iPhone app listens, transcribes on the device, and suggests questions while the meeting is still going. Afterward it writes a summary and a short quiz.

The hard part was cost. Every meeting calls an AI model over and over, so I layered limits per minute, per meeting and per person, and it degrades gently before it stops.

SwiftUI, Supabase, Claude.

Luma app promo showing three phone screens

Husbandmaxxing, on the App Store since September 22, 2026

An app that locks your phone until you show up for your wife

It blocks distracting apps at the system level, then unlocks them when you do something intentional. Levels, quests and a weekly report keep it going.

The hard part was Apple's Screen Time system, which needs four separate pieces to work together with the main app.

SwiftUI, Screen Time APIs, subscriptions.

See it on the App Store
Character screen with a rank ladder
Weekly insights screen with an attention score

Instant quote websites, demo with a made-up business

Type an address, get a price

The site finds the property from public map data, outlines the area on a satellite image, and prices the job while you watch. You can drag the outline if it's off.

One settings file turns the same template into a site for a different trade. There are 14 so far.

React, TypeScript, Leaflet, Cloudflare.

Try the live demo
Satellite map beside a measured area of 6,570 square feet and an instant price of 85 dollars
Sample address: a public park in Charlotte

Review requests, n8n, October 2026

An automation that asks at the right moment

When a job is finished, it writes a friendly message, waits two hours, and sends the customer a one-tap review link by text. No phone number on file? It emails instead.

n8n, Twilio, webhooks.

Job done
Write message
Wait 2 hours
Text
or email

Tools I use myself

Content agents

Five agents pull the week's data, find what worked, and draft the next plan. Scheduled weekly.

June 2026
Idea scout

Reads public forum posts and separates people asking for a solution from people selling one.

August 2026
Grocery planner

Matches our family list to weekly ads and confirmed store prices. A weekly Claude routine keeps it current.

October 2026
Honey Do

A private iPhone app and widget my wife and I use for requests. It syncs through iCloud.

October 2026

How I keep AI honest

Models are fast and sure of themselves. Each rule here came from something that went wrong, or could have.

The model proposes. A check decides.

Take One won't release a video that fails its quality gate, however confident the model was.

Prove the copy matches the original.

Moving Take One to the cloud changed its results in quiet ways. It took seven rounds of comparison to find why. Now a sync fails if one engine file differs.

Cap the cost before the bill does.

Luma limits AI use at several levels, so one unusual user can't run up the total.

No guessing.

My grocery planner once guessed a price and got it wrong. Now a price counts only after it's read from the store's page.

AI

Claude, Claude Code, n8n, Cursor, GitHub Copilot, agents, automation, evaluation and quality checks

Engineering

TypeScript, JavaScript, React, Angular, Node.js, Python, Java, Swift, AWS Lambda, PostgreSQL, DBT, Redis, REST APIs, Datadog, Splunk, Jest, Cypress, accessibility (WCAG)

Education

B.S. in Computer Science, Wright State University, 2016. Internships at Forte Industries (Swisslog) and Reynolds & Reynolds.

Got something slow, stuck or unclear?

LinkedIn