Eyewitness Sketch Lab

• 

  lobby, where player can choose difficulty and select suspects source

• 

  suspect glancing, with remain timer and filters if applied

• 

  suspect describing page by typing

• 

  alternative describe suspect method by conversation

• 

  sketch generation page

• 

  comparison page, AI analysis tells how well the player described

Inspiration

I got the idea from recent case about crown got stolen from louvre museum, those suspect was seen by many people walking by, but i was thinking how good people were for providing clues when it comes to describing the suspect's figure as i'm terrible to recognize faces.

What it does

The app/game retrieves a real suspect’s profile image from the FBI database via an API. The player is given X seconds to study the image before describing the suspect’s appearance from memory. Depending on the game mode, we can add visual noise or distortion to the image to further challenge the player’s observation skills. Once the player submits their description, Gemini is used to generate a sketch based on their input and then compare it to the original image. Gemini evaluates the similarity between the two images and provides a performance score reflecting how accurately the player described the suspect.

I use Gemini 3 Flash for High-speed multimodal vision comparison and safety auditing. i use Gemini 2.5 Flash Image: High-fidelity forensic rendering. i use Gemini 2.5 Flash Native Audio: Real-time voice-to-voice interrogation via the Live API.

How we built it

I used purely google AI studio, since it does not include a plan mode, i gave him a hackton rules in file and let it generate steps in another md file, once im satisfy with the "plan", i then ask him to execute.

Challenges we ran into

1. my account keep bugging when come to sync with github
2. can not revert only some file or only some line
3. the editor was so limited when comes to text research, file research, shortcuts etc...
4. AI studio try always to rewrite the entire file, i improved this by refactoring parts/components into other files

Accomplishments that we're proud of

A functional game within a little time

What we learned

the important of refactoring and well describing

What's next for Eyewitness Sketch Lab

Generate simulated camera footage of the suspect to further test the witness’s observation skills. Players can improve their ability to describe suspects by training across various formats and scenarios.

Built With

• fbi-wanted-api
• html5
• react
• tailwind
• tsx
• typescript