A smartphone with a dark screen lying on black stone beside a bare graphics card board, both edged in red light.

Dallas, Texas

Isaac Maple

I build AI that runs on the hardware people already own, and the networks that keep it safe.

  • AI Engineer, On-Device LLM Inference
  • Founder, TokForge
  • DMZ Network Architect, Bank of America

About me

Isaac Maple, smiling, in glasses, a dark cap and a black polo shirt, against a dark background.
Based in
Dallas, Texas
Now
Bank of America, TokForge.ai, Elevated-Marketing.io, Defcon-One
Education
University of Texas at Tyler
Certified
CompTIA A+, Cisco CCENT, SSCA (SIP)
Other
Eagle Scout

I’m Isaac Maple, an AI engineer and founder in Dallas, Texas. I build AI that runs on phones, and the servers and networks behind it. I spent 15 years keeping big networks running: carrier network operations centers, hospital networks, a 911 provider, and now the DMZ at Bank of America, where I design the secure zones between the internet and the bank.

In 2026 I built TokForge, an all-in-one on-device AI app for Android and iPhone: chat, vision, audio, documents, images and voice, with no account and no cloud. It has more than 10,000 installs on Google Play in 169 countries and a public iPhone beta, and it is published by Defcon-One, the company I have run since 2009. I am also building SurvivalOps, an offline survival and medical guide, and Folio, an AI game master for long-form interactive fiction.

I build the hardware side too. My AI server joins eight NVIDIA CMP 170HX cards to two RTX PRO 6000 Blackwell cards over 100 GbE and serves open models up to 1 trillion parameters, I publish the recipe as open source, and I run cmp.wiki, the community reference for these cards. At Elevated Marketing, the agency I co-founded, I built and run the hosting platform behind 20+ websites, our clients’ and our own, on a database cluster that spans four sites.

I measure before I claim. Every performance number on this site comes from a logged run on real hardware, on real phones and real servers. I care about software that works on the hardware people already own, infrastructure that fails safely, and honest measurement. Outside work I’m an Eagle Scout, and I like projects that help people when the network is down. That is where SurvivalOps came from.

By the numbers

  • 10,000+TokForge installs on Google Play, in 169 countries
  • 1Tlargest model served on my own hardware
  • 35Bparameter AI model running on a phone
  • 106model repositories on Hugging Face, 180,000+ downloads
  • 20+websites on a database cluster across four sites
  • 15 yearsin network engineering and security

What I build

See all eight projects
Built a per-chip routing layer in the Kotlin/C++ runtime that sends each model to the fastest backend proven stable on that chip (CPU, Vulkan or OpenCL GPU, or the Qualcomm Hexagon NPU), with the slowest proven path as the fallback on unknown hardware.
Status: Live

TokForge

Private, all-in-one AI on your phone. No account, no cloud.

What it is
An all-in-one AI app that runs entirely on your phone, on Android and iPhone.
Measured result
10,000+ installs in 169 countries and 3,000+ active users. One app, every modality, fully on-device: chat, photo, audio and video understanding, document search, image generation, speech, voice cloning and tool-calling agents. A 35-billion-parameter model runs on a phone at 9 to 12 tokens per second, and the phone’s NPU reads long prompts about 4x faster than its CPU.

Built with

  • Kotlin,
  • C++,
  • Swift,
  • Jetpack Compose,
  • SwiftUI,
  • llama.cpp,
  • MNN,
  • ggml,
  • Vulkan,
  • OpenCL,
  • SPIR-V,
  • Qualcomm QNN,
  • Core ML,
  • MLX,
  • TFLite/XNNPACK,
  • sd.cpp,
  • Kokoro,
  • Sherpa-ONNX,
  • WhisperKit,
  • Room,
  • Hilt
Two open black rack servers side by side, one with two graphics cards and one with a row of eight silver fanless cards, joined by a single cable glowing red.

Two servers, ten GPUs, one model

Server 1, stage 1

2 x RTX PRO 6000 Blackwell

96 GB each

Server 2, stages 2 to 5

8 x NVIDIA CMP 170HX

Supermicro AS-4124GS-TNR, 2 x AMD EPYC 7302, 64 GB of GPU memory

The full GLM-5.3 (753B) runs across both servers as one pipeline, at up to 104 tokens per second.
Status: Running

CMP 170HX AI server

Eight NVIDIA CMP 170HX cards in one 4U server, joined to two Blackwell cards over 100 GbE, serving models up to 1 trillion parameters.

What it is
A ten-GPU AI server I built: eight NVIDIA CMP 170HX cards in a Supermicro 4U chassis, joined to two RTX PRO 6000 Blackwell cards in a second server over 100 GbE RDMA.
Measured result
Mistral Large 4, a 1-trillion-parameter model, on the eight cards, and the full GLM-5.3 (753B) as one ten-GPU pipeline at up to 104 tokens per second. The recipe is open source, with my fixes offered upstream.

Built with

  • Supermicro AS-4124GS-TNR,
  • AMD EPYC 7302,
  • NVIDIA CMP 170HX,
  • RTX PRO 6000 Blackwell,
  • 100 GbE RDMA,
  • vLLM,
  • llama.cpp,
  • MTP,
  • PipeSpec,
  • NCCL,
  • VMware ESXi,
  • Linux
A fanless silver graphics card on a dark workbench beside the same card with its cover removed, a blank notebook and tools, joined by thin red lines like a diagram.
Status: Live

CMP 170HX wiki

cmp.wiki: the community reference for running LLMs on NVIDIA CMP 170HX cards.

Measured result
180+ articles, with data tables for 46 recipes, 24 platform reports and 63 confirmations, open as daily JSON snapshots and through a public MCP server for AI agents.
  • WordPress,
  • MediaWiki,
  • Cargo,
  • Page Forms
An old ledger lying open on a dark desk, its pages covered in red strokes, with a faint web of red light points rising above it.
Status: Pre-alpha

Folio

An AI game master for long-form interactive fiction that never loses track of your quests, gear or the people you met.

Measured result
A story ledger that tracks quests, inventory, NPCs and their secrets, events and backstory, recalled at 86% on a 152-question continuity benchmark: more than double the 40% from loading the whole story into context.
  • Python 3.12,
  • FastAPI,
  • PostgreSQL 16,
  • pgvector
A rugged phone in a rubber case on a creased topographic map over rock at dusk, a brass compass beside it and mountains behind.
Status: In development

SurvivalOps

A survival and medical expert in your pocket that works with no signal.

Measured result
Ask in plain words or snap a photo of a plant, an injury or a map and ask about it: a 4-billion-parameter vision model answers on the phone in seconds and cites the page it used.
  • Kotlin,
  • C++,
  • Python,
  • MNN
Three black server racks in a dark data center aisle, rows of red status lights behind perforated doors and cables in a tray above.
Status: Live

Elevated Marketing hosting platform

High-availability hosting for 20+ websites on a database cluster that spans four sites.

Measured result
A database cluster with five voting members across four sites. In a real 55-minute data center outage on October 29, 2026 it elected a new primary in about a minute, with no visitor-facing outage on load-balanced sites. A planned drill moved 20 domains, with writes tested, in 101 seconds.
  • Linux,
  • Apache,
  • PHP-FPM,
  • MySQL 8 Group Replication
A card payment terminal and a receipt printer with a curl of blank paper on a dark stone counter in a dim restaurant.
Status: In development

Elevated POS

A restaurant point-of-sale and ordering platform that never loses an order, even when the network goes down.

Measured result
Every sale is an event on an append-only log, so the system can always rebuild its exact state, and the floor keeps taking orders through an internet outage.
  • TypeScript,
  • Node,
  • Fastify,
  • WebSockets
A black workstation case with a tinted glass side showing two large graphics cards outlined in red light.
Status: Running

Home AI lab

A private AI datacenter in a rack: one router in front of every model, with multi-GPU serving, training and evaluation.

Measured result
One OpenAI-compatible endpoint for models from 27B to 753B, with hot swaps in seconds: a 320B model at up to 187 tokens per second and a 125B model at 185 tokens per second, each on two GPUs.
  • vLLM,
  • SGLang,
  • llama.cpp,
  • ExLlamaV3

Where I work now

Read the full resume
  1. DMZ Network Architect

    Bank of America

    Jun 2023 to PresentPlano, Texas (remote)

    I design and review DMZ and perimeter network architecture for internet-facing workloads, partner connectivity and secure access to banking systems.

  2. Founder and Developer

    TokForge.ai

    Feb 2026 to PresentRemote

    TokForge is an all-in-one on-device AI app for Android and iPhone: chat, photo, audio and video understanding, document search, image generation, speech in and out, voice cloning, tool-calling agents and web search, all running on the phone with no account and no cloud processing by default.

  3. Co-Founder and Technical Lead

    Elevated-Marketing.io

    Nov 2024 to PresentRemote

    I built and run the hosting platform behind 20+ websites, our clients’ and our own, for the digital marketing agency I co-founded: a database cluster with five voting members across four sites, AI-driven monitoring and layered security.

  4. Owner

    Defcon-One

    Jul 2009 to PresentDallas, Texas (remote)

    Defcon-One is my independent cybersecurity, AI and managed IT brand, and the publisher of TokForge.

Before that

  1. May 2022 to Jun 2023Senior Network Project EngineercentrexIT
  2. Mar 2021 to May 2022Director of Network Operations (promoted from NOC Manager)NGA 911
  3. Jul 2018 to Mar 2021Network Engineer IIISpectrum Enterprise
  4. Jul 2016 to Jul 2018Network Engineer IIIFrontier Communications
  5. Apr 2015 to Jun 2016NOC Engineer IBaylor Scott & White Health
  6. Sep 2014 to Apr 2015Telephony NOC Engineer ISuddenlink Communications
  7. 2008 to 2014Network operations and supportEarlier roles