• High-agency solo systems builder — basement lab, bare-metal Threadrippers, 100+ dockerized experiments, local-first by default.
• Allergic to theater — proof over applause. I publish the failure reports, the unvarnished evals, and the commit that deleted features my README invented.
• Harness & agent literacy — autonomous agentic loops, hierarchical memory tiers, sealed evals, budget routing, and tamper-proof verification rails so models can’t fake disk writes.
• Model surgery & local runtimes — open-weight inference stacks (vLLM, SGLang, Ollama), hybrid SSM architectures, steering, and abliteration on bare metal.
• Full creative & vertical stack — music production, latent-space generative art, interactive lab demos; alongside production engineering, SMB AI ops, and security-critical AuthZ/IDOR fixes.

• Agent Runtimes & Harness Engineering: Designing resilient multi-agent execution loops, memory compaction systems, tool verification rails, and recovery harnesses that don't fall apart under entropy.
• Open-Weights & Model Research: Red-teaming, steering, abliteration, benchmark stress testing, and real-world eval design on local GPU hardware.
• Stealth & Early-Stage Lab Acceleration: Acting as an autonomous player-coach who transforms raw research papers and whiteboards into robust, running containerized infrastructure on real hardware.
• High-Impact Missions & Open Source: Teaming up on ambitious tools, public-good AI infrastructure, or unconventional technical ventures with teams that value craft and radical candor.

• The Accidental Lobotomy — memory systems, green checks that lie
• MKUltra — agent OS thesis
• Refusal Lives in the Conv Blocks — Liquid LFM abliteration notes
• Art gallery · Music · Demos

• Open-source / open-weights labs and agent runtimes (Hermes-class, local models, evals, memory engines)
• Founders & teams who need a high-velocity builder to turn messy frontier architectures into solid metal
• Sovereign / self-hosted AI stacks, edge deployments, and unconventional technical products
• Causes and research projects that want an authentic builder-writer, not corporate slide-deck theater

• Roles that require hiding the lab notebook and only shipping pitch decks
• SEO content farms, synthetic benchmark chasing, and "AI wrapper" hype
• Teams that punish honesty about incomplete or failing systems

Whether you're developing an autonomous agent runtime, exploring novel model architectures, hacking on open-source tools, or untangling a gnarly production failure — let's connect.

Send me a note with what you're working on, why it matters, and where things get interesting. Repos, architecture diagrams, research papers, or broken production stories are always welcome.

Aurora / Denver · lab on metal · iceboks@duck.com