Home/Work/Case 07

AI Product
Enhancer

Reviews an AI product end to end from its demo video, its memo and its architecture, then writes the prompt that fixes it.

RoleArchitect & builder
WhenMar–Aug 2026 · Decasonic
StackGemini · OpenRouter · React
StatusLive
3inputs: video, memo, architecture
0–100score with a verdict
7scored sections
1ready-to-paste fix prompt
01 / The problem

Reviews that end
in a to-do list.

Product reviews usually produce opinions. Someone still has to turn them into changes, and in an AI-built product the person making the change is often another AI builder.

The firm also had no repository access to many of its own AI apps, so a normal code review was not possible.

02 / How it works

Watch it, read it,
then write the fix.

Demo videowatched by a multimodal model
Product memoparsed into value, audience, pricing
Architecturevia the prompt-copy method
Score0–100, seven sections
Suggestionsranked by severity
Fix promptpaste into your AI builder
03 / Decisions worth explaining

Why it’s
built this way.

01

A model that actually watches the demo

The demo video goes to Gemini for real multimodal analysis, compressed in the browser with retry and resume for large uploads.

02

The prompt-copy method

With no repo access, the app hands you a prompt to run inside your own AI builder and you paste the answer back. It became the most reused trick in the product.

03

A worldview you can edit

A knowledge base of the firm’s thesis, standards and design principles is injected into every evaluation, so scores reflect how the firm thinks, not generic taste.

04

Critique to fix in one tool

Pick the suggestions you want and it writes an implementation prompt with real code, ready for Replit, Lovable, Cursor or Bolt.

04 / What it produces

One score, seven
sections, one prompt.

  • A score out of 100 with a ring gauge and a bullish or bearish verdict, plus an executive summary.
  • Seven sections: video analysis, value proposition, features, architecture, UX, integration and data flow, and alignment with the firm.
  • Ranked suggestions, each with severity, category, reasoning and an action.
  • An async job pipeline with progress polling, so long videos never block the UI.

Need something like this? I build these end to end, as a full-time engineer, on contract, or with SKAI Group.

Book a 30-min call
Next case · 08Flora AI
Contact

Let's build

Need your content in Arabic, or an AI product built this carefully? One email is enough to start.