Dualite LinkedIn bio roaster: trace the prompt, model response and public app

Define the roast and the input boundaries

The 23 September 2025 walkthrough builds an AI-powered LinkedIn bio roaster in Dualite with Google Gemini. It enhances prompts with ChatGPT and adds rules before connecting Gemini Flash. Begin with the expected input and tone: the application should respond to the biography supplied, not invent personal information or turn a playful brief into unsupported claims about a person.

1. Use the retained builder route: https://alpha.dualite.dev/.
2. Follow project setup at 0:22 and prompt refinement at 2:00, reading added rules before accepting them.
3. Write what a useful roast should preserve from the input and what content should remain out of scope.

Connect the model and test the application response

The Gemini API connection begins at 3:02, debugging and live previews at 4:25, and testing at 6:12. These stages need separate checks: the interface can load while the model call fails, and a successful response can still ignore the intended tone or input. The description does not reproduce the exact prompt or credential configuration.

1. Follow the recorded API setup and keep credentials out of public files and screenshots.
2. Use a fictional test biography and compare the returned roast with the supplied facts and rules.
3. Inspect the failing action and response during debugging, then repeat the same case after the correction.
4. Check the final result at 10:38 for content and usable presentation, not only whether text appeared.

Deploy the intended version and keep bonus examples distinct

Pac-Man and a Gen Z finance app appear at 7:56 as bonus builds; Netlify and GitHub deployment starts at 12:04. The source retains the demonstration at https://bio-roaster-uh3i-alphaproject.netlify.app. That historical link is a reference, not a verified current uptime or quality claim.

1. Identify the reviewed repository and build before following the deployment sequence.
2. Open the deployed version independently and repeat the main input-to-response journey.
3. Keep each bonus project’s outcome separate from the roaster’s tests when judging Dualite’s usefulness.
4. Check present access terms before using code LUKEDUALITE; its inclusion in the old description does not confirm a current benefit.

The takeaway

A usable prompt-built app needs aligned rules, working model integration and a checked deployed journey.