מִיכָאֵל
INITIALISING fR3K

AI SYSTEMS ARCHITECT // NORTH QUEENSLAND, AUSTRALIA

fR3k

Independent AI systems architect & operator. I build autonomous systems where models, tools, infrastructure, hardware, safety and operations have to agree — and I prove them with tests, not prose.

14 active nodes
reticulum / lxmf mesh 98.7% link quality
systems nominal
See the Work github.com/fr3kchy ↗
// fR3k — system online (8s loop)

00 About

Independent AI systems architect and operator — callsign fr3k — working out of North Queensland. My surface spans local-first agent infrastructure, resilient communications, distributed control planes, model evaluation and optimisation, RAG, multimodal generation, embodied AI, and evidence-gated field systems.

I do not count scaffolds as finished systems, and I do not report simulated passes as execution evidence. A blocked gate is a valid engineering result.

01 Work on GitHub

Real repositories, grouped by the kind of problem they solve. Every one ships with executed evidence — tests, measured sweeps, or reproducible pipelines.

Resilient Communications

hermes-reticulum-platform

Independent Reticulum/LXMF transport plus a fail-closed Hermes policy gateway. Transport survives gateway failure.

38 tests

hermes-fleet-control

Secure local-first control plane for distributed Hermes nodes — HMAC heartbeats, durable task queue, authenticated events.

16 tests

Distributed Agent Operations

fr3k-infrastructure

Executable governance and evidence gates for community communications infrastructure. Checksum-verified, blocked-state honest.

evidence-gated

trajectory-replay-demo

Record and replay agent tool trajectories for model/prompt A/B testing. SQLite-backed comparative evaluation.

replay A/B

Local Model Engineering

qwen36-bench

Reproducible Ollama optimisation on constrained hardware — 35B MoE on 4 GB VRAM, 5.71 tok/s peak, all checks passed.

35B MoE · 4GB

ollama-local-rag-demo

Fully local RAG with Ollama and FAISS. Zero-cloud retrieval and generation path.

zero-cloud RAG

mcp-rag-pgvector-demo

Production-shaped RAG with pgvector, embeddings, MMR and evaluation. Reproducible harness.

pgvector RAG

Agent Evaluation

agent-eval-harness-demo

Golden-dataset agent evaluation and drift detection. 25-question baseline, local + hosted backends.

golden dataset

agent-cost-router-demo

Route tasks to the cheapest viable model instead of defaulting to frontier inference. Measured baseline.

cost routing

Domain Computation & Education

divine-arsenal

Domain computation joining ephemeris data, planetary hours and operational rules — deterministic Python, not symbolic prose.

deterministic

openclaw-workshop

Two-day hands-on curriculum for agents, MCP and AI coding tools. Structured labs + deployment material.

2-day labs

The reference names agent-core / reticulum-lxmf from an earlier mock aren't published — links above point to the live, verified equivalents.

02 How I Engineer

  • ⟁ Operational truth outranks mythology. Read the running system, not just its design document.
  • ⟁ Transport and control are separate failure domains. Optional AI must not weaken the underlying system.
  • ⟁ Evidence is graded. Observation, unit tests, integration tests and live acceptance are not interchangeable.
  • ⟁ Security boundaries are explicit. Inputs untrusted, queues bounded, privileged audit fails closed, cloud egress gated.
  • ⟁ Hardware claims require hardware. Virtual/LAN acceptance is never relabelled as RF, robotics or field acceptance.
  • ⟁ Useful failure reports beat polished fiction. A blocked gate is a valid engineering result.

03 Company

I also run MCP Intelligence — my company building psychology-first autonomous agent systems for clients (persistent memory, voice, orchestration, delivered as full source). That's the product side; this site is the engineering.

mcpintelligence.com.au ↗

05 Launch Blitz

The fR3k launch series — a 2-minute film plus vertical shorts for YouTube, Reels and TikTok. All rendered locally with HyperFrames.

// fR3k — engineering reel (45s)
Teasercoming online Systems Are Hardmodel · tools · ops Proof Not Prosetest it Launch Filmfr3k.dev live 35B / 4GBconstrained HW Resilient CommsReticulum/LXMF Agent Evalmeasure it Local-Firstzero-cloud RAG Blocked Is A Resultevidence-gated Inspect The Workread the code MCP Intelligenceclient systems Goldfishpersistent memory SOUL.mdbehavior is architecture Memorynot chat history Business Outcomecontinuity Full Sourceno lock-in Two-Week Buildcustom agent systems

04 Contact