CaseLens — Know What the Evidence Says. Know What to Check Next.
Inspiration
When something goes wrong at home, the evidence rarely lives in one place. A maintenance log records one date, a repair estimate lists another, and an incident note tells a slightly different story. All of these records may be sitting on a NAS, but understanding what actually happened still takes hours of manual review.
I wanted to explore a different kind of AI assistant: one that doesn't just summarize documents, but helps people make better decisions based on the evidence they already have.
That's why I built CaseLens.
What it does
CaseLens is a Uliya Skill designed to turn scattered NAS records into source-linked incident reviews with actionable next steps.
Instead of generating a single confident narrative, CaseLens follows a structured evidence-review workflow:
- Reconstruct the timeline: Distinguish when an incident occurred, was discovered, was reported, and was investigated.
- Trace the evidence: Connect factual claims to their source documents and avoid treating repeated statements as independent confirmation.
- Examine contradictions: Identify genuine discrepancies while distinguishing differences that may have reasonable explanations.
- Find what's missing: Highlight evidence gaps and unresolved questions.
- Recommend the next best action: Explain which document or clarification would be most useful to obtain next, and why.
For example, imagine reviewing records from a water-damage incident. Two repair estimates disagree by $700. Rather than simply flagging conflicting numbers, CaseLens directs the user to check whether the estimates cover the same work before drawing a conclusion.
The goal isn't just to explain what the records say. It's to help users understand what they should verify next.
How I built it
CaseLens is designed as a Markdown-based Uliya Skill using the official UGREEN Skill Creation Tool and development standards.
Its architecture orchestrates authorized NAS file discovery, document examination, evidence reconciliation, and structured incident reporting through supported Uliya capabilities.
The workflow includes explicit rules for source attribution, temporal reasoning, evidence relationships, uncertainty, permissions, and failure handling.
I used OpenAI Codex to assist with implementation, documentation, technical review, and validation.
The initial Skill package has passed the official static validation checks with zero errors and warnings. Runtime execution on a UGREEN NAS has not yet been verified.
Challenges and lessons learned
The biggest challenge was distinguishing useful AI reasoning from convincing but unsupported conclusions.
Different dates don't always mean contradictory records. Two documents repeating the same statement don't necessarily provide independent confirmation. And missing evidence shouldn't be replaced with a plausible explanation.
Designing CaseLens reinforced an important principle: a trustworthy AI assistant needs to know when the available evidence isn't enough.
Another challenge was keeping the workflow compatible with Uliya's supported capabilities without relying on custom executables, undocumented APIs, or connected surveillance cameras.
Accomplishments
- Designed a structured incident-reconstruction workflow rather than a generic document summarizer.
- Introduced evidence relationships, date-role reasoning, and context-aware cost comparison.
- Developed a Next Best Evidence workflow that connects unresolved questions to specific follow-up actions.
- Added explicit safeguards against fabricated sources, unsupported claims, and unauthorized file modifications.
- Completed official static validation of the initial Skill package.
What's next
The next milestone is validation on an actual UGREEN AI NAS.
I plan to test document retrieval, source attribution, permission handling, and repeated execution using realistic incident records.
I also want to compare CaseLens against conventional AI document summarization to evaluate whether its structured workflow meaningfully improves accuracy and the usefulness of follow-up recommendations.
Longer term, I envision CaseLens helping NAS users review equipment failures, maintenance issues, service disputes, and other situations where understanding the evidence matters more than producing another summary.
Built With
- codex
- markdown
- openai
- ugos
- ugos-skill-creation-tool
- ugreen-ai-nas
- uliya
- yaml
