Inspiration
Career guidance is usually presented as disconnected advice. Readers cannot easily see which claims support a recommendation, who owns a correction, or how a challenged passage should move through review. luzync turns that problem into a governed knowledge workflow rather than another chat answer generator.
What it does
A reader can choose a career, follow a continuous case through decisions and handoffs, and reopen the references and templates used in the work. The same Git-backed handbook is exposed to agents through a read-only MCP and a bounded route-career-handbook Skill.
In our controlled demonstration, a challenged W3C-backed passage is anchored to an exact target, content digest, and version. Local context first produces a review-required decision. Verified DataHub lineage then identifies an existing Git-owned review group, changes the route to join-existing-review, and suppresses a duplicate candidate.
How we built it
The public reader uses Astro with a Cloudflare Worker and D1 feedback API. Git-backed JSON content and evidence ledgers provide canonical IDs, digests, versions, and JSON Pointers. An Electron Studio gives operators a local governance view. Agents receive bounded context from the read-only luzync Handbook MCP and route-career-handbook Skill. DataHub v1.6.0 supplies read, search, and lineage context as a non-authoritative operational mirror. A controlled SDK and SQLite outbox carry identity-free overlay data, while a mutation-disabled DataHub MCP performs exact readback.
Challenges we ran into
The hardest problem was preserving one canonical authority while connecting reader content, operator governance, agent routing, Cloudflare feedback state, and DataHub lineage. Every decision had to stay bound to exact digest and version anchors. Stale digests, ambiguous lineage, unavailable DataHub, and failed readback must stop the workflow instead of inventing success. We also had to make the demo judgeable without exposing identities, raw feedback, secrets, local paths, or canonical writes.
Accomplishments that we are proud of
We delivered a reader-first handbook plus Studio, a read-only MCP, a bounded Skill, and a controlled D0 to D1 lineage demonstration. The qualifying exploratory slice covers 8 controlled pairs and 16 decisions with 0 hard failures in that bounded routing evaluation.
What we learned
Decision routing should be conditional: use local context when it is sufficient, query verified lineage only when needed, and fail closed on any invalid or unavailable dependency. Identity-safe overlays and exact readback make mirror integrations auditable without granting content authority.
What is next for luzync
We plan to add human-reviewed evidence refresh and accessibility validation, harden deployment and rollback checks, and expand only through controlled fixtures and explicit owner gates before considering any indexed or production release.
The evidence shown here is synthetic, controlled, and review-only. It does not claim user demand, content-quality improvement, production reliability, publication, or DataHub authority. Git-backed Handbook content remains authoritative, while DataHub is an operational lineage mirror.
Log in or sign up for Devpost to join the conversation.