The Judgemental Bestie

Icon
UI

Inspiration

Ever wished your best friend would just tell you the truth about your dating profile? We did. Between bland bios, cliché prompts, and questionable photos, we realized people need honest feedback — but with humor. That’s how The Judgmental Bestie was born: an AI friend who roasts you lovingly and helps you glow up your dating presence.

What it does

The Judgmental Bestie is an AI model that analyzes your dating profile (bio, photos, prompts) and delivers a brutally honest yet playful roast. It highlights what’s working, what’s cringe, and even suggests witty rewrites — all in the tone of your snarkiest best friend. Think of it as ChatGPT with sass and self-awareness.

How we built it

We built it using Claude Artifacts, leveraging Anthropic’s creative generation tools for structured tone and humor.

Core Model: Claude (via Artifact workflow) + fine-tuned prompt templates for controlled roast intensity
Pipeline: Each roast passes through tone filters and humor evaluators before being delivered — ensuring it’s funny but friendly

Claude Artifacts made iteration seamless — we could test prompt variants, refine tone, and visualize model responses in real time.

Challenges we ran into

Balancing humor vs offense — making the AI sassy, not savage
Training the model to maintain contextual awareness (so it doesn’t roast everything into oblivion).
Integrating LLM inference latency into a smooth real-time UI.
Filtering out biased or sensitive content while keeping the voice authentic.

Accomplishments that we're proud of

Built a functioning AI roast generator with personality and tone control. Integrated Claude Artifacts to iterate faster and more creatively. Achieved consistently humorous results that users actually enjoyed sharing.

What we learned

Humor is context-dependent — good prompt engineering beats fine-tuning for personality.
Claude’s contextual understanding makes creative outputs feel human.
A fun UX can turn feedback into engagement — people love being roasted safely.