Available for freelance work

I build things
that actually ship.

Freelance engineer specializing in web apps, AI integrations, automations, and rapid prototyping. I use the best tools available — including AI — to build faster and smarter.

Let's work together See my work

1// latest build
2const project = {
3  name: "client-dashboard",
4  stack: ["Next.js", "AI"],
5  shipped: true,
6  days: 4,
7}
8 
9// status
10console.log("✓ deployed")

6+Projects

iOS+WebPlatforms

AINative

ReactNext.jsSwiftNode.jsPythonClaude APIOllama

About me

Vibe-coding,
seriously.

I'm Jarren Patao — a freelance engineer who builds across the stack: AI voice assistants, iOS games, personal dashboards, distributed agent systems, and custom tooling. If it can be built, I'll figure out how.

My personal projects include Odin — a voice AI with its own agent pantheon — and Flask & Fury, a solo-built iOS roguelike. I work with whatever gets the job done: React, Swift, Python, Claude, Ollama, and real hardware integrations.

Available for freelance projects where the goal is to actually ship something — web apps, AI integrations, automations, or anything that needs a builder who can figure it out.

Web & Mobile

ReactNext.jsViteSwiftSwiftUITypeScript

Backend

Node.jsExpressPythonFlaskSwiftData

AI & Agents

Claude APIOllamaKokoro TTSOpenAI TTSWeb Speech APIsqlite-vec

Integrations

Google CalendarGmail APISpotify APISonosAPNsTailscale

What I build

Fast, focused, shipped.

I take on focused projects where speed and quality both matter. Here's what I do best.

Web Apps & Sites

Full-stack web applications and marketing sites — from landing pages to full SaaS products.

→Landing pages & portfolios
→SaaS dashboards & admin panels
→E-commerce & booking systems
→REST APIs & backend services

AI Integrations

Add AI capabilities to your product or workflow — chatbots, assistants, document analysis, and more.

→Custom AI chatbots
→Document & data extraction
→Content generation pipelines
→LLM-powered features

Automation & Scripts

Eliminate repetitive work with smart automations, bots, and custom scripts.

→Workflow automation (Zapier / Make / n8n)
→Web scraping & data pipelines
→Scheduled reporting & alerts
→API integrations & webhooks

Work

What I'm building.

Personal AI System

PANTHEON

A Norse-mythology agent system. Each god solves one problem. Odin coordinates them all.

Active development

Overview

One system.
Many gods.

Pantheon is a personal AI operating system built around Norse mythology. Each agent is named after a god and solves one specific problem — they orbit Odin, the central orchestrator, who routes tasks, manages memory, and coordinates action across your life. It runs self-hosted on a Mac Mini, reached from anywhere over Tailscale.

Odin · Central Orchestrator

Speak a command.
Odin routes it.

A voice-first, self-hosted AI OS. Chat inference runs locally on Ollama at zero API spend, with Claude held as an escalation path. Every turn is routed by tier, grounded in memory, and spoken back through a budget-capped voice pipeline to the web HUD, the native iOS app, Sonos, and Discord.

Voice In→Web Speech API · native iOS speech · “Hey Siri, talk to Odin”

Router→classifies the turn by tier and picks a target

Muninn→pulls only the memory this turn needs

Local Inference→qwen2.5 on Ollama · stable prompt prefix · first word in ~0.5 s

Tools & Agents→Calendar, Gmail, Brave Search, Sonos · Brokkr builds, Sindri researches

Voice Out→OpenAI TTS under a hard $5/mo cap → Kokoro local fallback

Shipped this quarter

JARVIS HUDNative iOS appSiri App IntentKokoro local TTSSpend ledgerSonos duck & resumePWA shell

GitHub

Muninn · Memory Layer

What Odin
remembers.

A three-layer vector memory system. Each request pulls only the context it needs — from a 100-token essence to atomic facts retrieved by cosine similarity. Embeddings are compressed 4× with TurboQuant (Google Research, ICLR 2026) and cached for sub-millisecond repeat queries. Retrieval now feeds every live chat turn, and every turn is archived back into the store.

Essence—Always-on ~100-token identity string — zero LLM calls, injected into every prompt.

Category Retrieval—Vector similarity over category embeddings — surfaces relevant domains, no LLM needed.

Atomic Facts—Top-k semantic search over 156 individual facts. Only fires when Layer 1 triggers.

SQLite + sqlite-vecOllamaTurboQuantLRU Cache

Sif · Meal Planner

What can you make
right now?

Tell Sif what's in your kitchen. Get three real meal options in under ten seconds. No sign-up, no subscriptions, no recipe blogs — just Claude and whatever's on your shelf.

List what you have

Type ingredients, no format required

Claude generates 3 options

Real meals, real ingredients, under 10s

Pick one and cook

No guesswork, no waste

React + ViteNode / ExpressClaude APIRailway

Víðarr · Daily Floor Tracker

Not your best days.
Just the floor.

Living alone removes ambient structure. Víðarr is a minimal nightly check-in — one question, three floors. It doesn't judge the ceiling. It just remembers whether you held the baseline.

01Eat something real

02Basic hygiene

03Move your body

React + ViteNode / ExpressRailway

Brokkr & Sindri · Build & Research Agents

The craftsman's
workshop.

Brokkr is Odin's build agent; Sindri is its research counterpart. Both run on local models via Ollama by default — no cloud dependency, no surprise bills. Next up: a durable milestone queue and a Tailscale executor so Odin can drive an unattended overnight build on a second machine and only wake you for roadmap decisions.

Build jobs

Brokkr runs a real tool loop — edits files, runs commands, reports back.

Research

Sindri gathers sources and drafts findings before Odin answers.

Budget gate

Build jobs escalate to the Claude API only under a hard nightly cap in the spend ledger.

Handoffs

Structured prompts and project memory so nothing is re-explained across sessions.

OllamaClaude APISpend LedgerNode.js

Also Building

ARIA

● In Dev

Amplitude Resolution Intelligence Agent — a conversational multi-tool AI agent built in Superblocks that investigates support tickets across Zendesk, Datadog, Linear, and Confluence autonomously. Compresses 20–40 min of manual research to under 5 min.

TypeScriptReactNode.jsSuperblocksZendeskDatadog

Claude Skills

● Shipped

Three Claude Code skills published org-wide: amp-week0-setup (dev env onboarding + troubleshooter), psp-account-context (Slack/Zendesk/Granola → Confluence synthesis), and customer-entitlements (Salesforce SOQL entitlement checker).

Claude APIMCPSalesforceZendeskConfluence

Sif

● Pantheon

Meal planner that asks what's in your kitchen and returns three real options in under ten seconds. Built for cooking alone with no inspiration.

ReactNode / ExpressClaude APIRailway

Víðarr

● Pantheon

A minimal nightly check-in tracker. Three floors: eat something real, basic hygiene, move your body. No streaks, no judgment — just the baseline.

ReactNode / ExpressRailway

Epoch Converter

● Live

Chrome extension for instant UTC ↔ local time conversion. One click, no site required.

Chrome ExtensionJavaScript

Code

How I work

From idea to live
in days, not months.

Discovery

We talk through what you need, what success looks like, and what's realistic. No fluff, just clarity.

Scope & Quote

You get a flat-rate quote with a clear scope. No hourly surprises, no scope creep without agreement.

Build & Iterate

I ship fast with regular check-ins. You see real progress early and can give feedback that matters.

Ship & Support

Deployed, documented, and handed off cleanly. Available for follow-up questions and future work.

Get in touch

Got a project
in mind?

I'm currently taking on new freelance projects. If you have an idea you want to build — let's talk. I'll get back to you within 24 hours.

jarrenpatao@gmail.com GitHub

I build thingsthat actually ship.

Vibe-coding,seriously.