Skip to content
View Foshowithit's full-sized avatar

Block or report Foshowithit

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Foshowithit/README.md

Adam Normandin

Agent runtimes · software factories · evaluations · local inference

I work on the execution side of AI: how an agent selects tools, performs work across processes and machines, recovers from failures, and proves what actually happened.

My main research project is RCOS (Recursive Capability Operating System). It connects goal routing, governed execution, artifact verification, and evaluation-gated capability reuse. The canonical codebase is private while publication and reproducibility checks are completed. Its central compounding hypothesis is still being tested.

Selected work

Project What's actually there
DSH Operator UI An operator view into agent work, capabilities, and execution receipts. The public, reproducible package targets an older DeepSeek Harness web profile; native Desktop integration is a development preview.
Shop OS — Pillow-Block Proof CAD artifacts, manufacturing-plan checks, a rejected toolpath plan, corrected advisory outputs, and runnable verification. A research proof, not production-ready machining instructions.
Jev × RCOS Routing Study Routing experiments with a cost ledger, adversarial cases, and a negative result for confidence gating.
FlowRouter A versioned manifest specification for portable executable capabilities and their evidence. Currently a spec, not a deployed capability network.
OW Workbench A proposed workflow-level evaluation system covering recovery, evidence, latency, cost, and long-horizon tasks. Pre-release; no official rankings.

Problems I'm working on

  • Software factories: from a repository issue to a reviewed, independently tested, reproducible release.
  • Evaluations: checking the actual deliverable, not taking the agent's completion message as evidence.
  • Capability reuse: measuring whether learned workflows improve outcomes against a fixed harness.
  • Owned inference: running and optimizing open models on hardware controlled by the operator or customer.
  • Portability: separating the agent runtime, permissions, workflows, and evidence so work isn't locked to one model or host.

Elsewhere

Optimized Workflow · Personal site

Pinned Loading

  1. dsh-operator-ui dsh-operator-ui Public

    Public RCOS operator integration: reproducible locked DSH web-profile package; native Desktop integration is a development preview.

    JavaScript 2

  2. ow-workbench ow-workbench Public

    Provisional: public AI workflow-execution benchmark (DESIGN/PRE-RELEASE, no official results yet)

    Python 1