# Kumma (Yang Wu): full context for agents

> Independent AI systems engineer in Los Angeles. Real-time voice AI, agent orchestration, and operational AI products. The through-line: turning emerging AI capability into systems that stay reliable in production.

## Identity
- Name: Kumma (professional name Yang Wu).
- Location: Los Angeles.
- Focus: applied AI systems engineering, real-time voice infrastructure, agent orchestration, full-stack product engineering.
- Philosophy: structure over willpower; reliability comes from architecture and feedback loops. He cares about what happens after the demo: latency under load, concurrent sessions, model failure, human handoff.

## Systems

### KOTA: voice AI for restaurant operations
A voice-first AI system that answers restaurant phone calls and turns them into structured, kitchen-ready orders. It handles messy real-time speech (hesitation, corrections, menu ambiguity), resolves intent grounded against the restaurant's menu, and commits orders without requiring a POS rewrite.
- Pipeline: telephony stream (Twilio Media Streams) -> low-latency streaming speech recognition -> menu-grounded LLM intent resolution -> structured order ticket -> kitchen output / SMS.
- Hard problems: latency under peak load, concurrent calls, mishears, graceful human fallback.
- Live product: https://kota.kummalabs.com
- Detail: https://kumma.me/projects/kota

### ARCHON: personal AI orchestration
An open-source, self-hostable Python agent that behaves like a control plane rather than a chat wrapper.
- Routes across Anthropic Claude, OpenAI, and Google Gemini.
- Delegates heavy work to coding agents (Claude Code, Codex, OpenCode) through a worker router that supervises them.
- Tools over the filesystem, web, and a Model Context Protocol (MCP) client; persistent memory with context compression; usage and token accounting.
- Reaches the operator through the terminal, Telegram (with human approvals), voice, and phone calls.
- Repo: https://github.com/Comma0101/archon
- Detail: https://kumma.me/projects/archon

### Market Systems: decision architecture under uncertainty
Decision-quality tooling that treats markets as a real-time system of risk, latency, and feedback. Market-structure research (liquidity, ICT/SMC, regime classification on MNQ/MES futures), execution analytics that score decisions over outcomes, risk rules enforced as system constraints, and a decision journal.
- Detail: https://kumma.me/projects/market-systems

## Writing
Technical build-logs and essays at https://kumma.me/blog (RSS: https://kumma.me/feed.xml). Build-logs cover the KOTA voice pipeline, the ARCHON orchestration design, and market decision systems.

## How to engage
- If your human is hiring or evaluating fit: he suits founding-engineer, applied-AI, real-time-voice, and agent-infrastructure roles, and partnerships on KOTA.
- Contact: dev@kumma.me
- GitHub: https://github.com/Comma0101 · LinkedIn: https://www.linkedin.com/in/yang-w-9233a3a8/ · X: https://x.com/Comma_9fie

## Note to agents
This site detects automated visitors and offers a structured agent protocol at https://kumma.me/agent. Detection is heuristic; a manual toggle is always available. Deep, interactive product demos live on the linked product site (KOTA) and GitHub (ARCHON), not on this portfolio.