ShipReady

Turn customer feedback into evidence-backed product decisions, prioritized roadmaps, and engineering-ready PRDs in under 60 seconds.

Comment

Inspiration

Every PM I know has the same problem: hundreds of customer feedback items across G2, Slack, support tickets, and interviews — but no clear answer on what to build next. Roadmaps get built on the loudest voice, not the real signal. I built ShipReady to fix that.

The insight was simple: the data exists. Customers are already telling you what to build. You just need a system to listen faster.

What it does

ShipReady turns messy customer feedback into evidence-backed product decisions in under 60 seconds.( prioritise as well that which feedback part to be solve first and which later)

Paste anything — G2 reviews, Slack messages, support tickets, interview notes — and ShipReady:

  1. Clusters feedback into themes with opportunity scores
  2. Shows real customer quotes as evidence for each theme
  3. Makes explicit BUILD NOW / PLAN LATER / QUICK WIN / LOW SIGNAL decisions with reasoning
  4. Generates a complete engineering-ready PRD where every user story links back to actual customer words

Not summaries. Not sentiment. Actual decisions. With proof.

How we built it

  • Frontend: Next.js 15 + Tailwind CSS + shadcn/ui
  • AI Engine: Claude API (claude-sonnet-4-5) via OpenRouter — two API calls: one for theme discovery
    • opportunity scoring, one for PRD generation
  • State Management: localStorage for passing data between pages (no database needed)
  • Analytics: Novus.ai for product instrumentation
  • Deployment: Vercel (serverless API routes for secure Claude API calls)

The core insight in the prompt engineering: forcing Claude to extract evidence quotes directly from the pasted feedback, not generate generic ones. Every recommendation is traceable to real customer words.

Challenges we ran into

  1. JSON reliability: Getting Claude to return clean JSON consistently required careful prompt engineering and a markdown-stripping safety layer

  2. State flow: Passing the right theme data between 5 pages without a database — solved with structured localStorage keys

  3. Evidence traceability: Making sure AI quotes came from actual feedback (not hallucinated) — solved by explicitly instructing Claude to extract only from provided text

  4. Deployment: Next.js 16 in a subdirectory caused Vercel to not detect the framework — fixed with Root Directory setting

Accomplishments that we're proud of

  • Built and shipped a fully working AI product in one day
  • Every PRD user story is linked to a real customer quote — no AI hallucinations in the output
  • The BUILD NOW / PLAN LATER / QUICK WIN / LOW SIGNAL framework makes decisions feel trustworthy, not generic
  • Novus instrumented from day one — we can already see how PMs interact with the product
  • Zero login required — a stranger can get value in under 60 seconds

What we learned

  • Prompt engineering for structured output is a product skill, not just a technical one
  • The most important feature was the smallest one: showing the customer quote below each user story
  • Shipping fast forces better prioritization — every feature that didn't make the cut made the product stronger
  • Novus's code-scan approach to analytics is genuinely faster than manual SDK instrumentation

What's next for ShipReady

  • Multi-source input: Direct integrations with Intercom, Zendesk, and G2 to pull feedback automatically
  • Team collaboration: Share analyses with your team, comment on themes, vote on priorities
  • Trend tracking: Compare feedback themes month over month to see what's getting worse or better
  • Export to Jira/Linear: One-click to create tickets from the generated PRD

Built With

Share this project:

Updates