Case Study UI / UX for AI 2026 General Assembly · Capstone

Synthread —
turning scattered feedback into decisions.

An AI tool that turns scattered, conflicting design feedback — from Figma, Slack, docs and meetings — into one clear, evidence-backed plan, while the designer stays in control.

RoleSole designer · Concept, IA, UI, prototype
ToolsFigma · HTML/CSS · Generative AI
DurationCapstone · 2026
DeliverablesConcept, persona, journey, hi-fi prototype, deck
Project prompt
Design an AI-powered product experience that earns user trust, balances automation with user control, and could scale across a team.

Four goals I kept coming back to.

An AI that organizes feedback only helps if people actually trust it. These four goals held that tension throughout.

01

Give designers their time back

Turn the busywork of collecting and reconciling feedback into the work of deciding.

02

Make the AI trustworthy

Every claim should be visible and sourced — never a black box.

03

Keep the human in control

The AI proposes with evidence; the designer always makes the call.

04

Scale consistently

One predictable model across tools and teammates — design-system thinking for AI output.

Designers spend more time collecting feedback than acting on it.

After a review, feedback lands everywhere at once — and the real work, deciding what to change, gets buried under the busywork of gathering it. Three frustrations show up every time.

Scattered

Across Figma, Slack, docs and meetings — none of it talking to each other.

Contradictory

“Make it bigger” vs. “tone it down,” and nothing flags the clash until it's too late.

Loudest wins

Decisions follow whoever spoke last, not the weight of the evidence.

Meet Maya — the designer this is really for.

Maya
Chen.

Maya Chen

Senior Product Designer · 6 yrs · Brooklyn

Maya leads two product areas and runs the weekly design review with PMs, engineers and a researcher. She wants to ship confident decisions fast — but loses her mornings reconciling feedback before she can act on any of it.

PrioritiesEfficiency · Detail · Evidence
Works withPMs · Engineers · Research
Tools dailyFigma · Slack · Notion · Dovetail
StyleDecides fast, defends with data
Goals
  • Ship confident decisions fast
  • Make sure feedback is addressed
  • Defend choices with evidence
Frustrations
  • Feedback scattered everywhere
  • Contradictions surface too late
  • Hours lost just reconciling

Where it breaks down today.

Mapping Maya's current flow showed exactly where a tool could help — her emotion bottoms out right at “make sense” and “decide.”

Scenario · Reconciling feedback after a design review
Step
1Review
2Hunt & gather
3Make sense
4Decide
5Act & defend
Quote
“Feedback's flying in from everywhere.”
“Now where did everyone put it?”
“What actually matters here?”
“What do I fix first?”
“Now I have to defend the call.”
Pain points
All at onceNo order, no priority.
Manual & scatteredChasing four tools.
Contradictions hiddenNothing flags clashes.
Loudest winsNot evidence-led.
Rationale lostHard to trace back.
Emotion
🙂
😕
😩
😟
😮‍💨

How might we…

Three questions that opened up the design space.

01
How might we turn scattered, conflicting feedback into a confident, evidence-backed decision?
02
How might we surface what matters — and where people already agree?
03
How might we let AI organize, while the designer stays the decision-maker?

Synthread connects your tools, does the synthesis, and leaves the decision to you.

Six things it does — and shows its work for each.

01

Ingest everything

Connect Figma, Slack, docs & meetings once — feedback flows in automatically.

02

Cluster by meaning

Group comments into clear themes by meaning, not keywords.

03

Detect conflicts

Flag contradictions and show both sides, side by side.

04

Surface consensus

Highlight where people already agree — instant low-risk wins.

05

Grade urgency

Rank by impact, frequency, stakeholder weight & evidence.

06

Cite real research

Ground recommendations in quoted, linked UX findings.

The design principle

The AI proposes with evidence.
The human decides.

Every claim shows its receipts.

Because the AI gives direction, it has to prove it. Two mechanisms carry the whole trust story.

Trust

Source chips & cited research.

Every theme, priority and suggestion carries a chip showing who said it and where, plus a confidence level — and recommendations quote real UX findings. Users never take the AI's word; they can always trace it back.

Graceful failure

“Needs context,” not guesses.

Feedback from docs and meetings has no pin location, so Synthread infers the screen from context — a doc's anchor, a transcript's timestamp, semantic matching. When it isn't sure, it doesn't guess: it flags the comment as low-confidence and asks the designer to assign it.

Designed as a system, not a screen.

A restrained, cool “tech” palette with one electric accent — and semantic colors that mean the same thing everywhere.

Type scale
Display
Heading
Body — clear and readable.
Caption · source chip
Semantic color
IndigoAI & primary
AmberConflict
TealConsensus
InkText

The whole flow, interactive.

Connect sources, see feedback pinned on the design, watch chaos become ranked themes, resolve a conflict with cited research, and leave with a prioritized plan. Tap around — it's live.

Walking through the flow

It starts on the synthesis view — 32 scattered comments organized into five ranked themes — and lets you explore every part of the product from the sidebar.

  • Sources — connect Figma, Slack, meetings & docs
  • Design — comments pinned to the actual screen
  • Synthesis — 32 comments become 5 ranked themes
  • Theme detail — conflict, cited research, options
  • “Needs context” — unplaced feedback, flagged not guessed
  • Action plan — prioritized, editable, traceable
Open full prototype

Where I'd take it with more time.

01
Test with real designers

Validate the conflict component — do the transparency cues actually land, or just look reassuring?

02
Onboarding & connect flow

Design the first-run experience of connecting sources for the first time.

03
Empty & error states

Build out low-confidence, no-evidence and failure states more fully.

04
Team-scale governance

Explore how the theme taxonomy stays consistent across a whole design team.

End of case study

Thanks for reading.
Let's talk.