Daisy LeungEvidence-first AI Workflow Architect
繁中Send a project brief
Menu
繁中Send a project brief

Selected public work

These projects show methods, fixtures, and build artefacts in public. They are grouped by capability line so you can see what kind of problem each one helps make inspectable.

Agent system / six capability areas06 / 06 mapped

This map names the part of agent work each public skill is designed to handle before the detailed evidence records below.

  1. 01 / ARCHITECTURE

    Agent Capability Architect

    Capability decomposition and mechanism architecture

  2. 02 / EVIDENCE

    Adaptive Multimodal Intelligence

    Multimodal evidence intake and sensemaking

  3. 03 / RELIANCE

    Reliance Compiler

    Claim-level reliance and verification planning

  4. 04 / REVIEW

    Repository Review Gate

    Repository validation and decision routing

  5. 05 / MAINTENANCE

    Codex Cleanliness

    Maintenance memory and regression detection

  6. 06 / SPECIALIST INTELLIGENCE

    HK Local Voice

    Specialist language and cultural judgment

Capability map, not a claim of complete coverage or one integrated production stack.

01 / Capability line

Trustworthy AI workflows

04
PUBLIC SOURCEFLAGSHIP

Reliance Compiler

A public skill for turning AI output into evidence-bound claims, assumptions, failure modes, and verification needs.

03 / RELIANCE
Claim-level reliance and verification planning
Why I designed it
I designed it because a plausible AI answer is not yet a bounded decision; its claims, assumptions, failure modes, and minimum remaining verification need to be explicit.

Evidence boundary
Its fixtures and local tooling demonstrate a method, not production reliability or a client outcome.

Read the case study
PUBLIC SOURCE

Repository Review Gate

A public Codex skill for validating repository findings and routing remediation through human-owned decisions.

04 / REVIEW
Repository validation and decision routing
Why I designed it
I designed it because a finding should be validated, explained, and routed to the right human decision before remediation begins.

Evidence boundary
The review is bounded by exercised repository evidence and does not establish that every repository is safe.

View public source
PUBLIC SOURCE

Codex Cleanliness

A public repository-maintenance skill that records a verified local-Git baseline and reports changes in findings.

05 / MAINTENANCE
Maintenance memory and regression detection
Why I designed it
I designed it so an agent can compare repository health over time against a verified baseline without hiding uncertainty or pretending to certify the whole repository.

Evidence boundary
It reports bounded maintenance evidence without certifying repository health, executing target code, or giving a numeric score.

View public source
PUBLIC SOURCE

Agent Capability Architect

An architecture-first public skill for deciding what an AI or agent project needs before building.

01 / ARCHITECTURE
Capability decomposition and mechanism architecture
Why I designed it
I designed it to decide what should become a skill, agent, workflow, tool, prompt, or no new component before a team builds more than it needs.

Evidence boundary
It demonstrates a decision method, not a client engagement, implementation result, or uniqueness claim.

View public source

02 / Capability line

Complex evidence and domain intelligence

02
PUBLIC SOURCE

Adaptive Multimodal Intelligence

An evidence-aware public skill for inspecting multimedia sources and reporting provenance, uncertainty, contradictions, and coverage.

02 / EVIDENCE
Multimodal evidence intake and sensemaking
Why I designed it
I designed it because inspecting video, audio, images, PDFs, and mixed sources is only useful when provenance, uncertainty, contradictions, and coverage remain visible.

Evidence boundary
Optional providers may be unavailable, and the project does not prove perception quality or client outcomes.

View public source
PUBLIC SOURCE

HK Local Voice

A public Hong Kong content skill that separates natural local wording, cultural currency, and contextual appropriateness.

06 / SPECIALIST INTELLIGENCE
Specialist language and cultural judgment
Why I designed it
I designed it because fluent Traditional Chinese is not the same as natural, current, and context-appropriate Hong Kong communication.

Evidence boundary
Its public anchors are not exhaustive; native review remains valuable for high-risk or community-specific work.

View public source

03 / Capability line

Product delivery

02
SOURCE PENDING

fanheungha Japan Travel Notebook

A Traditional-Chinese, mobile-first Japan travel notebook for organising trips, itineraries, packing, lists, and pre-departure checks.

Evidence boundary
The public repository identity is unresolved against the project registry, so this card is intentionally unlinked; it makes no deployment or user-outcome claim.

Public link withheld pending source-of-truth resolution
PUBLIC SOURCE

Global Work Atlas

A public static, local-first project-dashboard template for work across regions, priorities, analytics, and activity.

Evidence boundary
It is browser-local and account-free, not evidence of a hosted multi-user service or business outcome.

View public source

A useful first conversation

Start with non-sensitive context. We can identify the claims, review burden, decision owner, and evidence that would change the next step.

Send a project brief