Project case study

Shopping Assistant Insights

During my Home Depot internship, I built a local dashboard that grouped shopping-assistant conversations, surfaced unanswered needs and made the underlying exchanges easier to inspect. It gave a reviewer a starting point beyond reading logs one by one.

My role: Solo prototype direction and development with AI-assisted implementation.

Watch the interactive walkthrough

How it works

Load fictional conversations

A sample conversation set becomes a review queue.

The original prototype processes local logs. The portfolio uses an invented dataset with no customer information or external AI call.

Group the themes

Flag unanswered questions

Potentially unresolved exchanges rise to the surface.

Local checks assign unresolved and answer-risk flags. A reviewer still needs to inspect the exchange before treating the flag as a finding.

Compare and export

The review can move from an overview to a shareable table.

The dashboard compares a brand with its category and exports the analyzed rows to Excel. Demo counts illustrate the interface, not employer results.

Read the pattern, then inspect the conversation.

During my Home Depot internship, I built a dashboard for reviewing shopping-assistant logs. It grouped conversations into themes, added sentiment and answer-risk indicators, compared a brand with its category, and exported the review. Processing stayed local and did not call an external AI service.

Focus: Local rules, sentiment scoring, conversation flags and spreadsheet export
Status: Alpha prototype shared internally; adoption unverified

How it works

The local dashboard groups logs using rules, adds sentiment and unresolved-question flags, and supports brand/category comparison and spreadsheet export. It does not send those logs to an external AI service.

A decision that mattered

I treated a flag as a reason to investigate. A rule can identify a phrase or an unresolved exchange, but that does not make it a verified customer outcome. The dashboard needs to keep the underlying conversation available so someone can judge the context.

Status

I shared the alpha prototype internally for feedback. I do not have verified evidence of continued adoption, so this case study focuses on what I built and how it works.