00 / Introduction
I investigate hard technical problems and build systems around them — across software, infrastructure, AI tooling, research corpora, hardware concepts, and distributed computation. This site is a navigable record of that work: selected systems, open experiments, failed approaches, and the relationships between them.
01 / Currently
How software can perform research rather than merely retrieve it — evidence pipelines, claim structure, and verification that survives the chat session that produced a change.
That splits into active builds and exploratory threads — not a single product pitch, but a coherent investigation surface.
Aftermath verification receipts for coding agents, and Stackglass as the observability plane that feeds agents real state, failures, contracts, and history instead of guesses.
Distributed computation and polyglot execution through the Chimera / Parallax lineage — meshes, IR migration, and recovery control planes that treat infrastructure as evidence.
Also experimenting · Hardware-aware execution, local-first repository intelligence, and tactile/desktop interfaces where provenance and safety are defaults.
Sometimes that produces software. Sometimes research. Sometimes an experiment fails and still leaves a useful constraint — a scheduling assumption that collapsed, a metaphor that stalled without a runtime, a benchmark suite that was not ready to publish numbers.
Those outcomes belong here too. The Lab keeps unfinished and failed approaches visible when they produced knowledge. The polished systems on this page are not the whole story; they are the parts worth entering first.
02 / Selected systems
01 / Systems
Language boundaries are operational boundaries. Parallax treats migration as an IR problem: capture program state in one language, encode it as language-neutral representation, and restore or execute it elsewhere — so polyglot systems stop paying rewrite cost every time the runtime changes.
The work sits with Centralizer and Chimera in a broader computation stack: interop supervision, distributed execution, and evidence about what actually ran. Status is experimental — useful as architecture and machinery, not as a finished product claim.
02 / Compute
Most distributed-compute stories assume a rented control plane. Chimera asks what happens when peers discover each other, share work without a central broker, and keep storage honest as machines come and go — gossip membership, QUIC transport, Wasm-bounded jobs, content-addressed results.
It is a concrete Rust grid rather than a decentralization slide deck. Trust, incentives, and failure domains remain open research questions attached to running code. The interesting part is that those questions stay tied to machinery you can inspect.
03 / Agents
Coding agents write fluently and still guess about repository truth. Stackglass is an observability and verification layer for AI-assisted development: local-first evidence about state, tests, failures, configuration, contracts, and documentation — consumable by humans and by agents.
It pairs with Aftermath (durable verification receipts) and Cartographer / Arcframe (orientation and blast radius). The bias is simple: show me is cheaper than trust me. Fluency is abundant; evidence is not.
04 / The graph
Domains, lineage, and shared infrastructure form one knowledge graph — not a list of unrelated repositories. Agent tooling feeds verification planes; polyglot runtimes sit beside distributed meshes; hardware and science projects share the same bias toward provenance and inspectable state.
The graph powers constellation thinking, archive lineage, and related-work links on project pages. Open it when you want relationships instead of a flat index.
Open graph →05 / Archive
Selected systems above are the cinematic path. The archive is the dense index — curated metadata plus the GitHub cache — for when you need language, domain, and status filters rather than a guided tour.
06 / Work with me
Serious conversations usually start from a specific system, a research thread, or a portfolio-level diligence request. Prefer intent over a generic hello — engineering, research, collaboration, acquisition, or employment each map to different next steps.
Acquisition and licensing details stay understated on purpose. The work should establish value before transactional talk; `/acquire` exists for parties ready for that discussion.