Inspiration

Small businesses handle invoices, purchase orders, payment records and receipts every day. However, these records are often scattered across folders and different formats. Connecting them, checking amounts and identifying missing evidence can become slow, repetitive manual work.

I wanted to build a solution that goes beyond simply storing or summarising documents: a system that helps businesses understand how their transaction records relate to one another.

What WELLCOME does

WELLCOME (Business Evidence Intelligence) is a read-only AI Skill concept designed for UGREEN AI NAS.

It helps users search an approved folder, examine transaction documents and connect records when the available evidence supports the relationship.

Its central innovation is the Business Evidence Graph. Rather than making assumptions, WELLCOME labels findings as Supported, Consistent, Discrepancy, Not found in searched scope, Unreadable/incomplete or Possible match.

Every finding is linked to its source file path.

How I built it

I developed WELLCOME as a Markdown-based Skill package containing "SKILL.md" and four supporting reference files covering capabilities, evidence protocols, disclosures and validation.

I also built a separate browser companion application with a dashboard, evidence upload functionality and a Supabase data layer. Its AI assistant uses Gemini through a Supabase Edge Function, with Groq as an optional backup provider.

The browser application is a demonstration of the broader product concept. It is not integrated with or running on a UGREEN NAS.

What I learned

I learned the importance of distinguishing a plausible AI-generated answer from a conclusion supported by actual documents. Evidence traceability, uncertainty labels, controlled access and clear limitations are essential when dealing with business records.

Challenges and current limitations

The main challenge is ensuring that document relationships are supported by evidence rather than assumptions. Matching records must never depend on an amount alone.

The Skill package has been structurally checked, but its proposed UGREEN expert and action mappings still require verification with the official UGOS Skill Creation Tool. Physical NAS installation and testing have not yet been performed.

WELLCOME does not independently verify bank settlement, payment authenticity, fraud or legal validity.

The goal is to help small businesses move from scattered transaction documents to connected, traceable business evidence.

Our next priority is to validate WELLCOME against the official UGREEN UGOS Skill Creation Tool, confirm the supported expert capabilities and refine the orchestration workflow.

We plan to install and test the Skill in a UGREEN NAS environment, beginning with invoice-to-purchase-order reconciliation, receipt discovery and protection against malicious instructions embedded in documents.

Further work will focus on improving evidence matching, supporting more document formats, handling permission restrictions and large folders, and making the Business Evidence Graph more useful for small-business transaction reviews.

Our long-term goal is to make WELLCOME a reliable, transparent and evidence-driven business records assistant.

Built With

Share this project:

Updates

Submission history