Independent software studio
Meipath builds and runs iOS apps, web platforms, and the AI and agent systems behind them.
01· Studio
Meipath LLC builds and operates its own products, and takes on client engagements: design, code, infrastructure, and the operations after launch.
Engagements are scoped in quarters: six to twelve months, not weeks.
Founded and run by mize (opens in a new tab).
02· Architecture
Ride a build through the system below, itemize the platform it runs on, then dispatch the open-source team yourself. The same system ships our products, and this page.
A schematic of the meipath development architecture, drawn as the path a build travels. Intake: durable memory recalls prior work, skills load from a library, and tools connect over the Model Context Protocol. Dispatch: an orchestrator plans the run and fans it out to agents working in parallel, each in an isolated workspace. The gate: independent review plus an adversarial pass; a caught regression holds the change until every check passes. Ship: merge, continuous integration, deploy, live at the edge. The loop: what the run learned is written back to memory.
Platform
Six capability layers every build passes through, from frontier models to shipped software. Every claim here is true of the system that ships this page.
Frontier models, orchestrated together. Every change is drafted, then cross-examined by an independent model before it counts.
A composable library of engineering skills the agents load on demand: design, review, debugging, migration.
In-house Model Context Protocol servers that hand the agents the studio’s own systems: code, infrastructure, and knowledge.
Durable memory that compounds. Every session’s learnings persist and resurface in the next, on any machine.
Parallel agent swarms in isolated workspaces, each scoped to one task and gated before anything reaches main.
Autonomous loops that build, review, deploy, and watch production, then report the fleet’s activity back to the studio.
The fleet
An orchestrator dispatches each task to the specialists it needs, they work in parallel, and a quality gate passes before anything merges. Every agent on this map is a real, public repository.
the roster · 12 open-source agents
Dispatches the team (developer, reviewer, and adversarial passes), each in an isolated checkout, then gates and merges.
Turns a plan into a board of stories with dependencies and parallelizable dispatch waves.
Multi-agent research that auto-scales and returns confidence-graded findings.
Takes a repository to the studio standard: scaffold, conventions, review configuration, and a verification gate.
Plans JavaScript-to-TypeScript migrations in sequential waves.
Senior TypeScript review across types, architecture, security, and performance.
API design, review, debugging, optimization, and security.
CI/CD pipeline architecture, review, optimization, and security auditing.
Infrastructure agent: deploy, fix, and migrate, with guardrails on every change.
Senior iOS review: App Review simulation and engineering review, dual verdicts.
Apple UI craft: glass, springs, haptics, and accessibility.
Catches and blocks detectable AI patterns in prose, code, and UI.
03· Capabilities
SwiftUI and Swift 6 with strict concurrency. Live Activities, App Intents, widgets, plus the deeper platform integration each product requires. The native clients above are built this way.
TypeScript across the stack. Astro, Next.js, or Hono on the front. Postgres and Redis behind. WCAG 2.2 AA and Core Web Vitals as build gates. The hourly ingest platform above runs on this stack.
OpenAPI 3.1 contracts, idempotent writes, structured errors (RFC 9457). Event pipelines that ingest around the clock, with traces, metrics, and replay built in. The speech-to-text pipeline above uses the same traced, replayable design.
Speech-to-text, LLM features, and retrieval that ship inside real products. The live transcription pipeline above runs unattended. We evaluate the models before launch and monitor them in production.
Cloudflare at the edge, Railway behind it, GitHub Actions on a self-hosted runner fleet. Pinned dependencies, typed configuration, health-checked deploys with one-command rollback, plus internal services that monitor it in production.
Multi-agent engineering on the Model Context Protocol: in-house MCP servers and Claude Code plugins, adversarial review panels, and autonomous loops that build, review, and deploy. The same tooling builds and ships everything on this page.
25 public repos · github.com/TheMizeGuy (opens in a new tab)04· Work
Web platforms, iOS apps, and desktop clients: production systems, each shown with its stack.
Scanner audio in, timestamped transcripts out, unattended around the clock.
08:23 Had a real world call. They’ve shut off the crossing by for track work today.
14:45 Fire alarm engine 3. Fire alarm engine.
16:43 Virtual control from engine 1 4 1.
The live feed, map, and daily summaries as a SwiftUI client, built for iOS 18 with a Liquid Glass surface on iOS 26.
Berkshire County Scanner for iOSA collector fans every tracked realm into PostgreSQL behind a Redis cache, served as a public API and a dashboard.
bootybaybroker.comRealm heatmaps, price-history charts, and price alerts with Live Activities, widgets, and Siri shortcuts, on Swift 6 strict concurrency.
Booty Bay Broker for iOSA Rust tray app keeps an in-game addon stocked with hourly prices and returns player scans to the ingest service, closing the data loop.
Booty Bay Broker CompanionA Twitch live-stream chat source rendered beside embedded video, on Swift 6 strict concurrency.
App Store05· Process
A week with whatever exists (a codebase, a deck, or a first sketch) and the people behind it. A written assessment before we scope any work.
Two-week sprints with a Friday demo and a Monday plan. Tests written alongside the code, decisions documented inline.
Clean repos, runbooks, a transition plan. Ninety days of post-launch support, included.
06· Contact
We’re booking new engagements. Send a note about what you’re working on and we’ll reply within two business days.