AngeloDe LeonSoftware
& applied AI
San Francisco · Forward-deployed engineering · Applied AI

Angelo De Leon.Software. AI.Built to work.

I’m Angelo De Leon, a software and applied AI engineer in San Francisco and Head of AI at TechInSF Consulting. I work directly with client teams to scope the problem, build the integrations and software, deploy it, and train the people using it.

San Francisco, CA
Head of AI · TechInSF
Software · Applied AI
What I do

Every layer.
One curious engineer.

Customer workflows, working integrations, and software teams can use. Here’s the engineering behind that delivery.

Explore all services ↗
/ai-agents

Connect AI to the work.

I build integrations between AI assistants, browsers, and tools. BrowserHelm brings MCP, context handling, and configurable model routing together around a concrete automation problem.

MCPAgent toolingBrowser automation
Explore BrowserHelm →
/developer-tools

Make the workflow inspectable.

Useful systems need a way to understand what happened. Agent Watchtower brings events and session logs from multiple coding agents into a local database and dashboard, with clear measurement limits.

TypeScriptPythonSQLiteObservability
Explore Agent Watchtower →
/full-stack

Own the implementation.

Working with teams to turn a business need into a scoped implementation. I build across interfaces, APIs, and infrastructure, and carry the work through deployment and handoff.

Technical scopingReact & Node.jsAPIsDeployment
Explore SocialCanvas →
/applied-ai

Help teams use what ships.

I deploy local AI, build MCP servers and RAG applications, and help companies adopt Microsoft Copilot, Claude, ChatGPT, and Google Workspace Gemini. The work includes secure setup, employee onboarding, and practical training.

Local AI & MCPCustom apps & RAGCompany training
Explore AI services →
Latest work

Latest work.
Open source.

Tools I’m building, problems I’m exploring, and the code behind them.

01 / Agent infrastructure

BrowserHelm

A native macOS gateway between AI assistants and browser agents. Less context overhead. More deliberate automation.

BrowserHelm illustrated otter navigator at a brass and cyan helm with connected browser panels
02 / Developer observability

Agent Watchtower

A local observer for coding agents. Bring sessions, token use, and API-price estimates into one place.

Agent Watchtower illustrated raccoon lookout with conceptual agent timelines, token stacks and a character chart
03 / Full-stack product

SocialCanvas

An open-source Instagram widget platform. A way to bring social content into a website you control.

SocialCanvas illustrated red panda arranging floating photo tiles into grid, masonry and collage feeds
04 / Native macOS utility

Vibesomnia

Keep your Mac awake with the lid closed. Built for AI coding from a phone, remote browser tasks, and the night shift.

Vibesomnia illustrated banner: a bloodshot-eyed developer surrounded by glowing terminals at night
05 / Agent browser & systems integration

BrowserAI DUO

My independent BrowserOS Neo fork for browser agents: fresh scoped evidence, bounded actions, and a separate app and profile.

BrowserAI DUO cyberpunk command center concept artwork
06 / Browser observability

Claude Browser Accountant

A local Claude.ai extension for visible-text token estimates, API-equivalent prices and observed account-wide plan changes.

Claude Browser Accountant illustrated fox mascot reviewing a glowing ledger in a teal and amber night workspace
07 / Native macOS & applied AI

Claude Pocket Checker

A separate macOS panel for regular Claude Desktop Chat: local usage estimates, isolated account totals and before-send model advice.

Claude Pocket Checker illustrated owl advisor with a gold check card in an indigo night workspace
Out in the open

GitHub, since June 2026.

Day by day. Building, contributing, and sharing the work.

View my GitHub ↗

Contribution calendar from @RaDeleon.

Explore my contribution calendar on GitHub ↗.

As shown on my GitHub profile. Updated periodically.

Working with customer teams

Engineer.
Builder.
Still curious.

I’m Angelo, a software and applied AI engineer based in San Francisco. I’m focused on forward-deployed engineering: staying close to the customer’s problem, building across the whole system, and helping people use what we ship.

Explore latest work
01

Understand the workflow.

Work directly with the people who know the problem. Turn what they need into a scope the team can build and review.

02

Stay hands-on in the implementation.

Connect interfaces, APIs, AI tools, and infrastructure. Use debugging and observability to make the system understandable.

03

Carry it through handoff.

Deployment, documentation, and practical training belong to the same engagement. Help the team understand and use what was built.

Blog

What I’m building.
What I’m learning.

All posts
Native developer tools

Claude Pocket Checker: checking the model before sending the prompt

Building a separate SwiftUI macOS companion that estimates regular Claude Desktop Chat usage and gives local model and effort advice before sending, without modifying Claude.

Read the note
AI developer tools

Claude Browser Accountant: counting what the browser can actually see

Why I built a local Claude.ai extension for visible-text token estimates, API-equivalent cost and account-wide plan readings, with personal and work ledgers kept separate.

Read the note
Agentic browser & desktop

BrowserAI DUO beyond the browser: one agent for the whole job

My plan for BrowserAI DUO: an agentic browser and macOS desktop workspace, with phone control, remote workers, approved password-manager access, and Personal and Enterprise editions.

Read the note
Applied AI tools

Why I Built Overkill AI Check

Most people run their heaviest AI model on simple tasks. I built Overkill AI Check to show which Claude model and effort level a task actually needs.

Read the note
Native developer tools

Vibesomnia: keeping the AI night shift awake

Why I built Vibesomnia to keep my docked MacBook awake while I code from my phone with ChatGPT, Claude, and BrowserAI DUO, or leave tasks running overnight.

Read the note
AI agents & MCP

AI browser automation with MCP: the BrowserHelm design

How I designed BrowserHelm to separate AI planning from browser execution, recover compacted context, and verify automated work.

Read the note
AI developer tools

Local AI coding-agent observability: building Agent Watchtower

Why I built Agent Watchtower to inspect Claude Code, Codex, and Hermes activity, with local storage and explicit limits on time and cost estimates.

Read the note
Let’s build something

A good idea
starts with
a conversation.