InsightFlow: Evidence-Backed AI Research Repository

InsightFlow: Research decisions backed by evidence

Overview

InsightFlow is an independent product design project created for the Replit Buildathon. It brings transcripts, evidence, insights, themes, opportunities, and reports into one connected workflow.

The challenge

Research knowledge is often fragmented across recordings, notes, spreadsheets, and presentation decks. Teams lose the connection between what participants said, how an insight was formed, and why a product opportunity was prioritized.

The product direction

I designed a research repository where every conclusion can be traced back to evidence. The system connects interview transcripts and timestamped excerpts, supporting and contradicting evidence, AI-suggested and human-authored insights, themes and affinity-map clustering, prioritized opportunities, and shareable research reports.

Designing trust into AI

AI output is never presented as unquestionable truth. InsightFlow distinguishes Live AI from Demo synthesis, shows the origin of each item, preserves verbatim citations, exposes contradicting evidence, and keeps review and editing in human control.

Core experience

Researchers can move from interview to evidence, from evidence to insight, and from insight to opportunity without losing context. A visual theme canvas supports clustering, drag and drop, zoom, AI suggestions, and reset controls. Opportunity pages connect impact, evidence strength, supporting insights, and recommended next steps.

My role

Product strategy, UX research framing, information architecture, interaction design, UX writing, interface design, design system, responsive behavior, accessibility, and coded prototype direction.

Scope and outcome

The demo workspace includes 10 synthetic interviews, 6 evidence-backed insights, 6 themes, 5 opportunities, supporting and contradicting excerpts, and a complete evidence-to-report workflow. All research data is fictional and created for demonstration. Production metrics are not claimed.

What I would validate next

Comprehension of AI confidence and origin labels, speed and accuracy of evidence review, findability across large repositories, collaboration and permission models, keyboard and assistive-technology accessibility, and generated-report quality across research methods.

Built as a portfolio product study using React, TypeScript, responsive UI patterns, and AI-assisted synthesis with transparent fallback states.