AI / Software Engineer · U.S. Marine Corps Veteran

I build production AI & software, end to end.

AI engineer and Marine Corps veteran who designs, builds, secures, and ships real systems, not demos. A multi-agent AI platform with live users, a sports product in production, and an iOS app in beta. Below is the work, with the architecture you can dig into.

See the work ↓ Resume (PDF) Email me
Agents orchestration + routing RAG per-user knowledge base Tools approve-then-execute MCP-style connectors DevSecOps CI/CD + CASA
14
specialized agents under one routing layer
Live
deployed, multi-tenant, real beta users
100%
designed, built & secured by me
4.0
GPA, B.S. Information Systems & Cyber Security
Selected work

Projects

Shipped systems across agentic AI, full-stack product, and mobile. Each one designed, built, and deployed end to end by me.

Flagship · live
APEX — Multi-Agent AI Chief-of-Staff ↓

A production AI platform that runs a person's day. A routing layer sends each request to the right one of 14 specialized agents, which reason over the user's documents and live tools and draft every action for approval before it runs. Multi-tenant cloud, RAG, and a full DevSecOps pipeline, all built by me. Architecture below.

Node.jsSupabaseReactOpenAI / AnthropicRAGMCP-style toolsCASA Tier 2
Agents14, one routing layer
Action modelapprove-then-execute
Cloudmulti-tenant, encrypted
Statuslive, real beta users
Built byme, end to end
Royal Athletes drill with an animated court diagram Royal Athletes session builder
In production
Royal Athletes ↗

A full-stack sports training platform: four connected roles, 262 drills across 13 sports, with a drill engine that renders animated SVG diagrams from structured data. Web plus iPhone.

React Native (Expo)ReactSupabaseSVG animation
Surge iOS app: today's training plan Surge iOS app: progress dashboard
TestFlight beta
Surge — iOS training app

An athlete-path engine that retunes the whole program around the user across seven paths. Shipped to TestFlight with full account lifecycle, an App Store compliant paywall, and media handling.

React Native (Expo)In-app purchasesAuth lifecycle
Built within APEX
Content Reel Engine

A server-side pipeline that turns a structured brief into a finished branded vertical video: it plans the shots, renders them with ffmpeg, burns in captions, adds an ElevenLabs voiceover, and assembles the cut. Re-roll any single shot and it reassembles automatically.

Node.jsffmpegElevenLabs TTSAutomation
Flagship, in depth

Inside APEX

A user talks to APEX; a routing layer sends the request to the right specialized agent; the agent reasons over the user's documents and live tools and drafts any action for approval before it runs. I own the whole system: the orchestration, the RAG pipeline, the connectors, the cloud, the security, and the reliability.

APEX live interface: holographic head in a HUD with the agent dock and live telemetry
The live APEX interface, running. Voice-driven holographic HUD, the agent dock, and real-time telemetry.
01 · ORCHESTRATION

Routing + meta-agent

A certainty-gated router dispatches to 14 agents, with a strategic meta-agent that weighs the higher-leverage move. Distinct per-agent personas and server-side voices.

02 · SAFE ACTIONS

Approve-then-execute

Tool and function calling across Gmail, Calendar, and Slack. The model drafts every action; nothing runs without a human yes. Automation with a real safety gate.

03 · KNOWLEDGE

RAG document intake

PDF, Excel, and text intake into a per-user knowledge base the agents reason over, so answers cite the user's real world, not generic guesses.

04 · MULTI-TENANT CLOUD

Isolated per user

Node.js backend on Supabase with per-user runtime isolation, AES-256-GCM at-rest encryption, encrypted OAuth tokens, and self-serve password reset.

05 · DEVSECOPS

Shipped securely

GitHub Actions CI/CD with Trivy, Semgrep, gitleaks; per-request CSP nonces, SRI, SSRF and prompt-injection guards; a CASA Tier 2 review; actions pinned to commit SHAs.

06 · RELIABILITY

Kept healthy

A 4-state presence machine, health checks and keep-warm to kill cold-start gaps, voice retry/cache, prompt caching for latency and cost, and on-screen latency/throughput telemetry.

How it works

System architecture

Voice / Text user request, in ROUTING LAYER Certainty-gated + strategic meta-agent 14 SPECIALIZED AGENTS inbox calendar money content leads onboard real estate markets study memory coaching strategy each owns one job · per-agent persona + voice RAG knowledge base + Memory PDF / Excel / text · per user, encrypted APPROVE-THEN-EXECUTE Human-in-the-loop drafts queued for a yes CONNECTORS Gmail · Calendar Slack · Analytics MCP-style, OAuth, tokens encrypted FOUNDATION Node.js (ESM) · Supabase (Postgres/Auth) · React · Render · GitHub Actions CI/CD · AES-256-GCM at rest · CSP nonces · SSRF / prompt-injection guards · CASA Tier 2
APEX agent roster: the 14 specialized agents
The 14 specialized agents, each named and routable.
APEX connections panel: Gmail, Calendar, Slack and more
Per-user connectors: Gmail, Calendar, Slack, and more, over OAuth.
Under the hood

Tech stack

AI / LLM
Multi-agent orchestrationTool / function callingRAGPrompt engineeringPrompt cachingLLM evaluationMCP-style connectorsOpenAIAnthropic (Claude)ElevenLabs
Backend
Node.js (ESM)REST APIsSupabase (Postgres / Auth)Multi-tenant isolationStripeBrevo
Frontend
ReactReact Native (Expo)JavaScript (ES2022)HTML / CSSWeb Audio / Canvas
Cloud / DevOps
RenderNetlifyGitHub Actions CI/CDDockerHealth checks / keep-warmTelemetry
Security
AES-256-GCM at restOAuthCSP nonces + SRISSRF / CSRF guardsPrompt-injection hardeningTrivy · Semgrep · gitleaksCASA Tier 2 (DAST)
Learning
TypeScript (strengthening)Python (strengthening)
Who I am

Veteran. Builder. Finisher.

I am an AI and software engineer and a U.S. Marine Corps infantry veteran. I learn a domain by building in it, and I take a thing from blank page to deployed and keep it healthy after launch. APEX exists because I wanted to prove that one person could design, build, secure, and operate a real multi-agent AI system. I work daily in Claude Code, Cursor, and Copilot, and I am completing a B.S. in Information Systems and Cyber Security with a 4.0 GPA.

Ownership, urgency, and finding a way when the path is not paved are not slogans to me. They are how I operate.

Let's talk.

I am open to AI / software engineering roles. The fastest way to judge me is to read the architecture above and the projects.

andy@andyakotia.dev Resume (PDF)