Applied AI Engineer

Jared Casillas

I build production AI systems with Claude.

Apps that ship, not tutorials. I design, build, and run AI products end-to-end — from the prompt architecture to the service worker. real-time voice · prompt caching · OAuth · serverless · based in Mexico (US time zones)

Book a call contact@jaredcasillas.com

85production releases on one PWA

40+serverless functions live

3products in the wild

0frameworks required

Selected work

Three products. All live.
All built solo, end-to-end.

Real deployments with real users — an AI companion in daily production, a client site running in Alaska, and a business I founded.

Try the live demo ↗

Flagship · AI companion PWA

Second Brain

A personal AI companion that runs as an installable PWA — chat, voice, push notifications, and a living knowledge graph, all orchestrated through the Claude API.

Prompt caching on every turn — large system context cached via the Anthropic API, cutting token cost and time-to-first-token on repeat calls.
40+ Netlify serverless functions behind one app — Claude orchestration, Google OAuth (Gmail + Calendar), Web Push, ElevenLabs voice, Netlify Blobs persistence.
Offline-first, 85 shipped releases — hand-rolled service worker plus a custom canvas knowledge graph with multi-pass glow rendering tuned for 60fps on iPhone.

Open the live menu ↗

Client work · Sitka, Alaska

Cold Water — digital menu

A QR menu for a working bar & grill in Alaska — scanned at the table, readable instantly, even on slow northern connections.

One self-contained HTML file — no framework, no build step, tiny payload; first paint is effectively instant on weak connections.
QR-first UX — guests land on a venue selector and are reading the menu in two taps, tuned for one-handed use at the table.
Custom coastal identity in pure CSS + inline SVG — animated waves and an illustrated wordmark with zero image downloads.

Visit the live site ↗

Founder project · 3D printing

CD PRINTS

The commercial site for a 3D-printing service aimed at architecture students — designed to turn a model file into a quote request.

Print-shop identity system — a signature CMYK bar and a scroll-driven “printing…” progress meter that mirrors a real 3D print.
Conversion-focused structure — gallery, process, and quote CTA with IntersectionObserver reveal choreography.
Complete SEO / OG layer — meta descriptions, social cards, and link previews ship with the page.

How I work

One call. One price.
Shipped async.

01 — CALL

30 minutes, scoped live

You describe the problem; I tell you honestly whether AI helps, what it costs to run, and what I'd build. No deck, no discovery phase.

02 — FIXED-PRICE BUILD

One number, agreed upfront

Fixed scope, fixed price, fixed deadline — settled before work starts. You watch progress on a live URL from day one, not in a repo you have to run.

03 — ASYNC DELIVERY

Loom updates, US hours

Milestone walkthroughs on video, decisions in writing, releases you can tap on your phone. I work from Mexico in US time zones — overlap, not overnight lag.

Stack, by role

What I actually run in production

Orchestration

Claude API · Anthropic
Prompt caching · cost + latency
Tool use / agent loops
Structured outputs

Frontend

Installable PWAs
Service workers · offline-first
Canvas rendering · 60fps
Vanilla JS · zero-framework

Infra

Netlify serverless functions
Netlify Blobs · persistence
Edge deploys · instant rollback
Cost-aware architecture

Integrations

OAuth 2.0 · Gmail / Calendar
Web Push notifications
Voice synthesis · ElevenLabs
MCP-style tool servers