Forward Deployed AI Engineer · Founder · Builds & Runs Local AI Models
Richard Echols
I don't just talk about AI. I ship it, and I build and run my own local AI models on my own hardware. 26 paying customers. $1.6M managed. One engineer.
Forward Deployed AI Engineer · Founder · Builds & Runs Local AI Models
I don't just talk about AI. I ship it, and I build and run my own local AI models on my own hardware. 26 paying customers. $1.6M managed. One engineer.
26+
Paying Customers
$1.6M+
ARR Managed
10,500+
Lines Shipped
USPTO
Trademark
What I Built
A production AI business operating system with 26+ paying customers and a USPTO trademark. Not a side project. Not a demo. A real product people pay for.
“I didn't just learn AI — I built a business with it.”
10,500+ lines of production code. Direct LLM API integration with prompt caching. Multi-agent orchestration. Cross-platform distribution. AI-powered customer support. Built and shipped by one person.
Production API integration, not a wrapper. Custom prompt caching delivers 90% cost reduction.
4 AI personas with distinct specializations operating autonomously on real business tasks.
File management, web search, code execution, scheduling, email, and more built into the engine.
Production-grade auth layer on Cloudflare Workers. Token tracking, usage limits, billing logic.
Native installers for macOS (ARM64) and Windows (x64). Automated CI/CD build pipeline.
10+ cron jobs handling support, monitoring, content moderation, and customer notifications.
What I Built
My own locally-trained AI model. A 35.9B-total, roughly 3B-active mixture-of-experts coding model, LoRA fine-tuned on Apple Silicon with MLX and published Apache-2.0 on Hugging Face. I build and run my own local AI models on my own hardware.
“I build and run my own local AI models on my own hardware.”
Kaiju-Coder is a business-niche coding model, tuned for the work small businesses actually need: building websites, wiring up Stripe, generating invoices, handling CRM and intake, and shipping automations. It runs entirely local through Ollama, LM Studio, and opencode.
Honest framing: this is a scoped, business-niche model, not a frontier model. It is built to be useful and fully owned, not to top a leaderboard.
Trained, fine-tuned, and served on my own private compute fleet. The whole pipeline, from data to weights to local serving, runs on hardware I own and control.
35.9B total parameters with roughly 3B active per token. A sparse MoE built for fast, local inference.
Adapted on Apple Silicon using Apple's MLX framework. The training and tuning ran on my own machines.
Published openly under Apache-2.0. Anyone can download the weights and run them.
Served through Ollama, LM Studio, and opencode. No frontier API call required to use it.
Scoped for websites, Stripe, invoices, CRM and intake, and everyday automations.
Trained and served on private hardware: two Mac Studios (incl. a 256GB M3 Ultra) and two NVIDIA DGX Spark GB10 units.
Enterprise Impact
I bring the same intensity to enterprise accounts that I bring to my own products. The numbers speak for themselves.
$1.6M+
ARR Portfolio
Full Premier book, embedded across 5 enterprise and government accounts
96%
Gross Retention
110% net retention with expansions
6/6
On-Time Renewals
Plus 2 account expansions
25%
Usage Increase
Weekly active usage growth across accounts
2025 - Present
IBM (Apptio)
2024 - 2025
IRS
2023 - 2024
VT Industries
2015 - 2023
County Government (8+ years)
Projects
Real products. Real users. Real code. Every project here either has paying customers or is in active development heading there.
AI business operating system with 26+ paying customers. Full CLI with 14 native tools, multi-agent orchestration, and prompt caching.
Native macOS desktop client built with Swift and SwiftUI. Direct API integration with the Kiyomi engine, native notifications, and system-level shortcuts.
4 AI personas running concurrently via tmux sessions, coordinated through Telegram. Each persona specializes in different business operations.
Gamified fitness iOS app that turns workouts into RPG quests. Character progression, achievement system, and social challenges.
Technical Proof
These aren't resume keywords. This is what I use to build and ship production software every day.
Watch Me Build
I don't just claim I build things. I record myself doing it and put it on the internet. Live coding, product launches, and real engineering.
How to Install Kiyomi Max on Mac
Claude Can Now Control Your Computer — Kiyomi Max Demo
Credentials
MS Mathematics
Walden University
BS Technology Education
Georgia State University
CompTIA Security+
TBM Executive Certification
Cloud FinOps Certified Practitioner
Public Trust
DoD Security Clearance
Interactive
Don't read a PDF. Ask questions and get instant answers about my experience, skills, and qualifications.
Resume Assistant
Powered by Richard's actual resume
Try: "Does he have cloud experience?" or "What makes him stand out?"
Contact
I'm open to AI platform engineering roles, technical consulting, and interesting collaboration. Reach out.
Download Resume