Customer-facing AI implementation: discover the real workflow, design a bounded system, evaluate it, ship with proof, and hand off operable infrastructure. Outcome: Portfolio provides an 8-minute review path linking 9 live studios, each with clone-and-run demo proof.
Mike Rodgers — Forward Deployed Engineer (PDF)
This repository is the review path for hiring managers and technical leaders evaluating Forward Deployed / AI Solutions / Applied AI engineering work.
It is not a dump of experiments. It shows the operating layer that decides whether an AI workflow survives production: requirements, context boundaries, evals, auditability, rollout, and customer handoff.
Screening question answered here:
Can this person turn an ambiguous AI mandate into a supportable deployment with evidence?
- This README
docs/implementation-playbook.mddocs/architecture.mddocs/evaluation-strategy.mddocs/customer-discovery.mddocs/public-boundary.md- Live studios linked below — clone one and run a demo
Most AI projects do not fail because a model cannot produce text.
They fail because the implementation never becomes trustworthy inside the customer workflow.
This portfolio encodes the implementation layer I care about:
- translating messy business context into usable system boundaries
- designing agentic workflows with human approval and rollback
- building eval loops that catch regressions before users do
- making outputs inspectable with audit logs and provenance
- turning one-off delivery into reusable templates and tools
workflow reality
→ bound context / tools / memory / policy
→ deterministic gates
→ agent steps (earned autonomy)
→ eval harness + human approval
→ ProofPacket / audit log
→ operable handoff
| Studio | What it proves |
|---|---|
| proof-studio | Signed completion claims; false-done catch |
| jake-studio | Operator OS + L10 harness + closed loops |
| mesh-studio | Multi-node subsystem probe / boot / recover |
| agency-studio | Role contracts — Builder ≠ Verifier |
| app-factory-studio | Spec → scaffold with definition-of-done chain |
| communications-studio | Gated communication protocol engine |
| strategy-studio | Deterministic strategy routing |
| doctrine | Rules agents load before they act |
| Path | What it shows |
|---|---|
docs/implementation-playbook.md |
Discovery → go-live phases |
docs/architecture.md |
Reference architecture for grounded AI workflows |
docs/evaluation-strategy.md |
Contract tests, workflow evals, false-done plants |
docs/customer-discovery.md |
Questions that turn vague asks into buildable scope |
docs/public-boundary.md |
What is intentionally excluded from public surfaces |
templates/customer-implementation-plan.md |
Reusable engagement plan skeleton |
templates/go-live-checklist.md |
Practical go-live checklist |
examples/audit-log/ |
Audit log pattern notes |
examples/eval-harness/ |
Eval harness pattern notes |
This repo is documentation-first. The fastest executable proof on the account:
git clone https://github.com/mrodgersjs-web/proof-studio.git
cd proof-studio/packages/rigforge
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
rigforge demo- Outcome first.
- Ground everything.
- Evals before scale.
- Human approval where risk is real.
- Auditability is product quality.
- Smallest useful loop, then harden.
No customer PII, prospect lists, auth cookies, ToS-risk engagement automation, or internal client monorepos.
See docs/public-boundary.md.
- Script:
docs/video-script.md - Recording:
assets/demo.mp4(pending render)
- Profile: mrodgersjs-web
- proof-studio · jake-studio · doctrine
MIT — see LICENSE.
| Practice | Here |
|---|---|
| Employer summary | top of README |
| 60s / smoke proof | |
| Public boundary | |
| Claim under test | '"playbooks + review path present"' |
| Related fleet | profile · resume · patents teaser |
If fails, the README claim is considered false until fixed.
- Preview:
assets/demo.gif
