ClauseLock

Inspiration

Right before a client call, an insurance agent faces a messy pile of files—brochures, manuals, FAQs, claim forms. Keyword search finds files; cloud chat invents answers or demands uploads. We wanted a moment where the NAS calmly says: here’s the clause, here’s the path—or we don’t know.

What it does

ClauseLock runs on UGREEN AI NAS as a Skill for that pre-call scramble:

1. Auto-sort authorized insurance files into a tree: line of business → product → doc type (manual / FAQ / comparison / brochure / other), with rename suggestions.
2. Ask for confirmation before move/overwrite; uncertain files go to unsorted.
3. Rely on UGOS Intelligent Learning for local semantic indexing (no custom indexer in the Skill).
4. Answer product/coverage questions only with file path + verbatim quote; refuse when evidence is missing; list conflicts instead of picking sides.
5. Optionally save the Q&A + citations as a note for the meeting.

How we built it

Official UGOS Skill Creation Tool / Uliya standard: SKILL.md orchestrates file, document, and knowledge expert agents. Taxonomy and “cite-or-refuse” rules live in the Skill; platform idle-time learning is the retrieval base. Inspired by real insurance knowledge-hub workflows, rebuilt within hackathon capability limits—no custom scripts or external RAG services in the package.

Challenges we ran into

Mapping a deep product taxonomy onto supported NAS file ops; keeping AI enrichment strictly bound to retrieved snippets; designing clear confirm/refuse paths for Round 1 without device runtime proof.

Accomplishments that we're proud of

A scene-first Skill (agent before the call), not a generic RAG demo; deepest-leaf filing + human confirm; evidence-locked answers that match insurance’s “show me the clause” culture.

What we learned

Vertical beats generic: judges remember the client-call panic more than “another local RAG.” On AI NAS, the win is auditable local workflow, not reinventing the index engine.

What's next for ClauseLock

On Finals hardware: end-to-end sort → cited Q&A demo video; tighten multi-version policy conflicts; optional claims “missing-docs checklist”; prefer local models for fully offline answering.

Built With

• document-q&a
• expert-agents
• knowledge-base
• local-ai
• markdown
• rag
• skill.md
• uliya