Trellis
Portfolio sketchUncommon Schools — Senior SWE, AI Solutions

An AI platform for mission-driven education orgs — declaratively configured, DLP-aware, audited by design.

Trellis is a sketch of the JD's hardest bullet: “reusable AI stack templates and low-code tools that empower non-technical team members.” Every skill, policy, PII rule, and integration is a YAML manifest in platform/. The UI is the rendered state of the repo. A change is a PR. A rollback is git revert.

Acting as Teacherthe catalog and the audit visibility re-render when you switch.stub mode (no API key)

Skills

5

platform/skills/*.yaml

Personas

4

platform/personas/

Bindings

4

platform/policies/

DLP rules

3

platform/dlp/

Connectors

3

platform/connectors/

Everything is a manifest

Skills, personas, policies, DLP rules, connectors — all files inplatform/. Validated by zod at startup; the server refuses to boot if a manifest is invalid.

DLP on every invocation

Regex + similarity + classifier layers run on both input and output. Yellow banners for warnings, hard-block on restricted PII. The skill manifest names which rules are required.

Audited by design

Every invocation is logged with persona, DLP findings, and outcome. Admins open the /audit tab and see the last 200 calls; non-admins get 403.

Mapped to the JD

reusable AI stack templates

platform/skills/*.yaml

low-code tools for non-technical staff

/skills + the manifest PR workflow

library of Prompt Blueprints

blueprints/ + owner/reviewers fields in skill manifests

100% compliance with PII audits

platform/dlp/*.yaml + lib/dlp.ts enforcement

secure API integrations with core databases

platform/connectors/*.yaml with scopes_allowed/forbidden

documentation of AI environments and data flows

the manifest tree IS the documentation