The problem
Tasting sheets were paper, documents stalled in approval without anyone noticing, and any analysis meant digging through the pile by hand.
What I built
An end-to-end ASP.NET Core MVC system: open a tasting session, pick lots from ERP production codes, score, run a three-step signed approval, then analyse in a seven-tab report with an “Ask AI” chat that answers in charts.
Key features
- Four test families, up to five tasters and three approvers per test
- 1–5 scoring grid with comments and pass/fail, built for tablets
- Three-step approval with signature images; each step emails the next approver
- Controlled PDF report, plus an audit copy that can be edited without touching the original
- Seven-tab analysis report that cross-filters in the browser
- Ask-AI chat in Thai or English that draws charts in its answers and exports to Excel
Under the hood
- People, product and customer names become codes before anything reaches the LLM, and are restored in the streamed answer, so real data never leaves the company
- Answers stream over SSE from OpenRouter with a model fallback chain, logging tokens and latency against quota
- One compact column-based snapshot lets the browser cross-filter every tab without extra API calls