Careem AI Challenge · Working prototype

Design Feedback Copilot

Paste design notes for any Careem flow. Get a senior-level, severity-ranked usability review in seconds, with a concrete fix and rewritten copy for each issue.

Built by Ghanesh RS
Principal Product Designer
www.ghanesh.com
Food checkout Rides Payments Groceries
Prototype note: this demo runs an on-device heuristic engine so you can try it instantly. In production it calls an LLM with the prompt shown lower on this page. Tip: click a flow to load a sample.
Your prioritized review will appear here.

The idea

In a super app like Careem, one design org ships across rides, food, groceries, and payments. Every screen competes for the same attention and trust, and review time is the bottleneck on quality. Design Feedback Copilot gives every designer a fast, consistent first pass before human critique. It reads a flow, checks it against the lenses a senior designer would use, and returns ranked issues with concrete fixes and rewritten copy. Humans keep the judgment calls; the tool clears the basics and raises the floor.

The production prompt

The shipped version runs this against an LLM. It is also runnable as-is in ChatGPT, Claude, or Gemini for a live result.

You are a Senior Product Designer running a design review for a super app
(rides, food, groceries, payments) on mobile. I will give you design notes,
a screen description, or UI copy. Return a prioritized usability critique.

Review each flow against these lenses: clarity, cognitive load, information
hierarchy, error prevention and recovery, accessibility, trust and consent,
localization (Arabic and right-to-left), and mobile-first constraints.

For every issue output:
- Severity: Blocker, High, Medium, or Low
- Heuristic: the principle it violates
- Problem: the specific issue in this flow, one sentence
- Fix: a concrete change, one or two sentences. Rewrite weak UI copy inline.

Rank by severity, then user impact. Be specific to the described flow, no
generic advice. End with the three highest-leverage changes and one metric
you would watch to confirm the fix worked.

Design notes:
[paste here]

How it scales at Careem

One designer gets a faster loop. A design org gets a shared quality bar. The same engine plugs into a Figma plugin for in-canvas reviews, a pull-request check on shipped copy, and an onboarding aid that teaches the house heuristics. Guardrails matter: it flags and suggests, it never auto-changes design, and every suggestion names the heuristic and a metric so the team can verify rather than trust blindly. It is explicitly localization-aware, since Careem ships Arabic and right-to-left first for many markets.

100-word summary

Design Feedback Copilot turns rough design notes into a prioritized, severity-ranked usability review, written like a senior designer. It checks each flow against clarity, cognitive load, hierarchy, error recovery, accessibility, trust, and Arabic right-to-left localization, then returns specific fixes and rewritten UI copy, ending with the three highest-leverage changes and a metric to watch. For a super app like Careem, where one team ships across rides, food, and payments, it gives every designer a fast, consistent first pass before human critique, cutting review cycles and raising the quality floor. This prototype runs offline; production calls an LLM with the prompt above.

Prototype by Ghanesh RS for the Careem Senior Product Designer AI challenge. The interactive demo uses a transparent heuristic engine; the production design uses the LLM prompt included above. No confidential data is used.