Real-time voice · agent systems · Los Angeles

I build AI that runs in production, not in demos.

Real-time voice, agent orchestration, and decision systems built around messy inputs, visible guardrails, and reliable outputs.

  • KOTAcall -> order
  • ARCHONroute -> worker
  • Marketssignal -> rule
See the systems →Start a conversation

Open to founding-engineer, applied-AI, and real-time-voice work, plus KOTA partnerships. Usually replies within a day.

Operational proof

Systems that convert messy inputs into reliable action.

live voice systemKOTAcall -> ordermessy restaurant phone speech -> kitchen-ready order ticketclarifies ambiguity before commit (opens in new tab)open-source control planeARCHONroute -> workermulti-step development task -> inspectable agent work sessiontrace, policy, recoverydecision-quality toolingMarket Systemssignal -> rulenoisy market context -> scored decision recordprocess over P&Lengineering notesField Notesbuild -> explainsystem implementation -> reusable operating insightno fabricated metrics

I build systems where intelligence becomes action, not just output.

  • AI systems
  • real-time voice
  • product engineering
  • agent orchestration
  • markets
  • visual practice

Selected work

The systems I am building.

01

KOTA

Turns restaurant phone calls into structured, actionable orders without requiring the restaurant to replace or deeply integrate its POS.

  • real-time voice
  • LLM orchestration
  • menu grounding

live voice/order flow

Input
messy phone speech
Transform
streaming STT + menu grounding
Output
kitchen-ready order
Guardrail
clarify before commit
Live demo
deployed
Real-time
speech to order
POS-independent
no replacement
Enter chapter →
01 / KOTA
callaudiotokensintent · menuorder ticketOrange chickenx2Chow meinx1confidencehigh
02

ARCHON

Unifies multiple AI models and coding agents into one orchestration layer for real development workflows.

  • orchestration
  • coding workers
  • active R&D

open-source control plane

Input
multi-step dev task
Transform
models + tools + workers
Output
inspectable agent session
Guardrail
trace, policy, recovery
3
models orchestrated
3
coding agents delegated
Open
source, self-hostable
Enter chapter →
02 / ARCHON
workerstoolsmemorymodelschannelssafetycoordinatorroute, trace, recover
03

Market Systems

Converts discretionary market reading into a rules-based MNQ/MES execution framework designed to prioritize process, risk control, and repeatability over daily P&L.

  • research
  • risk
  • execution

decision-quality workflow

Input
noisy market context
Transform
setup taxonomy + risk rules
Output
scored decision record
Guardrail
process over P&L
MNQ · MES
futures studied
Decisions
scored over PnL
Rule-based
execution framework
Enter chapter →
03 / Market Systems
calmvolatilerisk
04

Experiments & Visual

Interactive web, 3D, photography, and writing. Visual and written research that feeds how I design systems.

  • 3D
  • photography
  • essays
  • interface
View studies →Read essays →
  • Structure over willpower.Reliable outcomes come from architecture and feedback loops, not motivation.
  • Autonomy.Building toward control over what I work on and how.
  • Systems before labels.The work is studying complex systems and building tools to operate inside them.

Building something difficult? Let's examine the system.

For founders, technical leaders, and collaborators. Tell me the problem and the constraint.

Emaildev@kumma.me
GitHub LinkedIn X

A short brief is enough to start.

agent protocol