AI systems builder shipping end to end: AI agents do the work, a human gate on every merge. Live systems: CoreWise (turns videos and articles into actionable insights across multiple AI models), Truenote (cited knowledge answers for customer-service teams; answers with citations or declines), willaicite (free deterministic audit of whether AI answer engines can retrieve and quote your page), and KineFractal (systematic trading). Guides and process notes at corewise.academy.
This repo is the flagship piece: a public, auditable agentic software-development pipeline and the audit tool it exists to build.
One scored engineering report per repo in the portfolio. Browse the visual viewer at audit.corewise.academy. Most of the portfolio is private: the audit tool and pipeline are public and deterministic, and the published reports give a scored read on repos you cannot open (reproducible by the owner from the pinned commit).
Read the scores with this in mind: the tool counts only what is recorded on GitHub. Every repo here is solo, and my reviews almost always run as handoff audits in separate AI sessions, which leave no GitHub artifact. That link is the actual skill: it drafts a self-contained audit prompt for a zero-context session and charges it to falsify the work, not approve it. The deterministic scorer cannot see those sessions and gives them no score credit, so the review dimensions understate my actual practice. Most reports carry that practice as an owner attestation outside the scored evidence, and every report lists what the tool could not see.
| Repo | Score | Grade | Report |
|---|---|---|---|
| ryanportfolio/ryanportfolio | 78/100 | Strong | report |
| ryanportfolio/Truenote | 71.2/100 | Developing | report |
| ryanportfolio/Corewise.Academy (private) | 69/100 | Developing | report |
| ryanportfolio/AI-Firmware | 68.6/100 | Developing | report |
| ryanportfolio/PixelSwarm (private) | 68.4/100 | Developing | report |
| ryanportfolio/Local-CPU-only-PTT | 64.9/100 | Developing | report |
| ryanportfolio/range (private) | 56.6/100 | Early | report |
| ryanportfolio/Extract-Video-Wisdom (private) | 55.7/100 | Early | report |
| ryanportfolio/githelp (private) | 52.1/100 | Early | report |
An agentic-SDLC audit tool: point it at any GitHub repo and get a deterministic, scored report on AI-agent development discipline. Scored dimensions include agent-vs-human commit ratio, PR review coverage and review-catch rate, human merge-gate presence, CI test/eval gate presence and pass history, batch size, commit-to-merge lead time, plan-before-code evidence, and audit-trail completeness, rolled up DORA-style. The scoring core is deterministic: same repo state, same score, no LLM calls in the scoring path.
Every change since the bootstrap commit is built exclusively through the pipeline it documents. Each one flows:
flowchart LR
A["Plan note"] --> B["Agent build"]
B --> C["Independent AI review<br/>(fresh context, adversarial)"]
C --> D["CI test + eval gate"]
D --> E["Owner-authorized merge"]
- The building agent never grades its own work: the reviewing agent gets fresh context and an adversarial prompt (refute, don't approve).
- Merge authority stays with the owner. Merges are agent-executed under a standing, session-scoped owner authorization, revocable at any time. That is disclosed, not dressed up as a manual click; details in governance.
- Every PR links a plan note, passes the gate, and carries its review trail.
The public PR history of this repo is the living demo. Solo project, zero external users. The pitch is publicly auditable process, a reusable framework, and verified portfolio evidence, not adoption.
| Path | Purpose |
|---|---|
app/ |
The audit tool: scoring engine, collector, CLI, report renderer, site generator (TypeScript / Node 20, GitHub API). |
.github/workflows/ |
The pipeline itself: test+eval gate, AI-reviewer template. Written to be copy-pastable into other repos. Viewer deploys via Vercel (vercel.json). |
governance/ |
Human-in-the-loop checkpoint map, audit-trail contents, NIST AI RMF mapping, merge-execution disclosure. |
plans/ |
One plan note per PR, the plan-before-code evidence this tool scores. |
reports/ |
Published fleet audit reports, live. One per audited repo. Each report was owner-approved before publication. |
playbook.md |
How to run this pipeline on any repo. |