Agentic AI Systems
Multi-agent, self-adversarial systems on the ReAct loop. Reasoners that plan, act, and coordinate — shipping outcomes, not chatbots.
METIS · KaziAI · GenJob▄▄▄ ███▄ ▄███▓ ▒█████ ██████ ▒████▄ ▓██▒▀█▀ ██▒▒██▒ ██▒▒██ ▒ ▒██ ▀█▄ ▓██ ▓██░▒██░ ██▒░ ▓██▄ ░██▄▄▄▄██ ▒██ ▒██ ▒██ ██░ ▒ ██▒ ▓█ ▓██▒▒██▒ ░██▒░ ████▓▒░▒██████▒▒ ▒▒ ▓▒█░░ ▒░ ░ ░░ ▒░▒░▒░ ▒ ▒▓▒ ▒ ░
SR. SOFTWARE ENGINEER — NAIROBI, KE
Multi-agent AI · payment-grade security · scalable backends.
Most engineers either build or break. I do both — that's the whole point.
I architect autonomous, multi-agent AI systems and the security that keeps them honest: offensive tooling, exactly-once payment integrity, and platform backends proven at scale. 1,715 commits across public and private work say it better than a tagline.
Multi-agent, self-adversarial systems with the ReAct loop — METIS, KaziAI, GenJob. Reasoners that plan, act, and coordinate at runtime.
Offensive tooling and defensive integrity. A 10-round audit that surfaced 22 bugs incl. criticals — HTML injection, broken auth gates.
Django/FastAPI/Go services at scale — a 12.6M-record portal, real-time marketplaces, and a +60% throughput rewrite in Go.
Verifiable credentials and trust graphs — LinkedTrust: Hyperledger Aries, AnonCreds, W3C VCs, the AT Protocol.
Hire me for the hard parts — the autonomous systems, the security that proves they're safe, and the backends that don't fall over. Outcomes, not demos.
Multi-agent, self-adversarial systems on the ReAct loop. Reasoners that plan, act, and coordinate — shipping outcomes, not chatbots.
METIS · KaziAI · GenJobOffensive testing and defensive integrity. I find the holes — HTML injection, broken auth gates, payment races — before attackers do.
22 criticals found · MpesaGuardDjango / FastAPI / Go services that hold under load. A 12.6M-record portal, real-time marketplaces, and a +60% throughput rewrite.
Arizona Sunshine · GoVerdeDockerized, self-hosted, single-command deploys. From repo to production without the drama — and the monitoring to keep it up.
Docker · VPS · CIAn opportunity & risk-intelligence platform for Arizona — real client work, in production. Two products run on it:
A campaign-finance transparency portal serving 12.6M records — searchable, fast, Django + React.
Offensive-security multi-agent AI — self-adversarial, scope-gated agents that recon, plan, and exploit within authorized bounds.
A real-time errands marketplace — FastAPI + WebSockets + escrow, React/TS front-end, live order state.
Exactly-once payment integrity for M-Pesa / Daraja — idempotency, reconciliation, no double-charges under failure.
Network monitoring + intrusion awareness — watches the local network, flags rogue devices and anomalies.
+ Tensor Marketplace (Stripe) · Street-Well SACCO audit · LinkedTrust verifiable credentials — github.com/Emos21 ↗
Started shipping on GitHub. Foundations in Python and full-stack web — the years that became 1,715 authored commits.
Real-time marketplaces and product builds — Tensor (Stripe), GoVerde (FastAPI, WebSockets, escrow). React/Next front-ends on Python/Go back-ends.
Arizona Sunshine portal serving 12.6M records; a Go-microservices rewrite for +60% throughput. Backends that hold.
Offensive and defensive work — WiFi Sentinel, MpesaGuard payment integrity, a Street-Well audit that surfaced 22 bugs. Built KaziAI, a ReAct multi-agent platform.
Building METIS — a multi-agent, self-adversarial, scope-gated security platform. Recognized as a senior engineer. ▮ currently active
Python · Go · TypeScript · JavaScript · SQL
FastAPI · Django · Go microservices · REST · GraphQL · WebSockets
React · Next.js · Vue · Tailwind
PostgreSQL · MongoDB · Redis · pipelines
Multi-agent · ReAct · LLM orchestration · self-adversarial loops
Docker · Compose · self-hosted VPS · single-command deploys
Go on — try to break it. Type help to look around, or cat about.txt.
prefer async? email me →
gh: Emos21 · in: amos-mwangi