TeamPulse

Inspiration

Teams can leave a planning meeting with different ideas about the project’s purpose, priorities, and handoffs. TeamPulse helps uncover those gaps early, without grading teammates.

What it does

Teams paste named project materials, then members complete short, private text interviews independently. Questions begin broadly and adapt to each person’s responsibilities. TeamPulse supports two independent checkpoints: Idea Alignment when scope is being finalized, and Final Readiness before submission or judging.

When answers appear to conflict, TeamPulse asks neutral follow-up questions first. Findings link to excerpts from the project materials and saved responses. They distinguish Supported, Contradicted, and Unverified claims; missing evidence is not treated as proof of a defect. Members can challenge findings, propose actions, and revisit them later.

How we built it

The two-page web app uses HTML, CSS, and JavaScript, with IndexedDB saving progress in the current browser. A Node.js and Express server handles validation and AI requests. Together AI’s openai/gpt-oss-120b model generates topics, interview questions, clarifications, and findings. Zod validates model output; the server also checks that quoted evidence appears in the supplied records.

Challenges

Early questions focused too narrowly on implementation details. We changed interviews to start with the project’s purpose, then probe shared responsibilities, individual contributions, and handoffs. We also added neutral clarification before conflict findings.

AI can produce convincing but inaccurate quotes. TeamPulse checks evidence against the full saved sources and keeps a review incomplete when a quote cannot be verified.

What we’re proud of

• Independent interviews conceal earlier answers from the next respondent in the shared-device flow.
• Findings are evidence-linked, challengeable, and actionable.
• Progress stays in the browser, so an AI request failure does not erase completed responses.
• The app uses no accounts and gives no scores or rankings.

What we learned

AI findings need clear boundaries: a project claim is not proof of implementation, and missing evidence is not proof of failure. Team members also need ways to correct questions and challenge conclusions.

What’s next

We want to test question quality with real teams, improve transcript review and topic assignments, and make checkpoint history easier to revisit.

Built With

• cheerio
• css
• express.js
• html
• javascript
• mammoth
• multer
• openai
• pdf-parse
• yauzl
• zod