Lets Roast Bolt

analysis
view submissions
Feed page

Inspiration

What it does

letsroastbolt.today is an anonymous, AI-powered feedback platform with a twist. Here's how it works:

Roast, Don't Report: Users visit the site and anonymously submit their feedback about the bolt.new platform in the form of a witty "roast."

Get a Witty Reply: Instantly, our AI persona, "The Roast Master," analyzes the submission and fires back a humorous, in-character comeback, making the user feel heard and entertained.

Intelligent Analysis: Behind the scenes, a powerful AI engine deconstructs the roast, ignoring the sarcasm to pinpoint the core message. It assigns it one of 14 specific categories (like Editor Experience or Build & Deployment) and one of 7 nuanced sentiments (like Frustrated or Constructive).

Radical Transparency: All of this structured data is aggregated and visualized on a public /analysis dashboard, allowing anyone to see real-time trends, top issues, and the overall community sentiment.

How we built it

This project was architected with a "prompt-first" philosophy. The core intelligence isn't buried in complex code, but rather lives within a suite of carefully engineered natural language prompts that guide a Large Language Model (LLM).

Our system uses a dual-prompt AI architecture:

Prompt 1: The Core Analysis Engine: A clinical, analytical prompt whose sole job is to receive the raw user text and translate it into a perfectly structured JSON object with our custom categories and sentiments.

Prompt 2: The Roast Master's Comeback: A creative, personality-driven prompt that takes the original text and the category from the first prompt to generate the witty, context-aware reply that the user sees.

This AI core is supported by a modern web stack, including a serverless backend to handle the API calls, a PostgreSQL database to store and query the structured results, and a responsive JavaScript frontend to deliver the seamless user experience.

Challenges we ran into

The Persona Paradox: Our biggest challenge was teaching a "Roast Master," a persona designed for wit and sarcasm, how to handle genuinely positive feedback. Early versions would respond sarcastically to praise, which felt wrong. We solved this by refining the prompt to make the persona comically surprised and flustered by kindness, which kept the character intact while still validating the user's compliment.

Achieving Nuance: Distinguishing between a Negative and a truly Frustrated user, or a Feature Request from an Inquisitive question, required significant prompt iteration. We had to provide the AI with very detailed definitions and examples for each category and sentiment to achieve the level of accuracy we wanted.

Data Integrity: Ensuring the AI always returned perfectly formatted JSON without any extra conversational text was a technical hurdle solved through rigorous instruction in the prompt, telling it to output only the raw JSON object and nothing else.

Accomplishments that we're proud of

The Dual-Prompt Architecture: We're incredibly proud of designing a system with two specialized prompts that separate the analytical "brain" from the creative "voice." This makes the system robust, scalable, and easy to debug.

A Genuinely Fun UX: We succeeded in turning the chore of giving feedback into an enjoyable and memorable interaction. The engagement we see is a testament to the power of a strong, well-executed theme.

The Transparency Dashboard: Building a system that not only collects feedback but also analyzes and presents it back to the community in real-time is a major accomplishment that fosters trust and accountability.

What we learned

Prompt Engineering is Product Design: Our biggest takeaway is that crafting the prompts wasn't just a technical task—it was the primary way we designed the product's logic and user experience. The words in the prompt directly define the behavior and personality of the application.

A Strong Persona Drives Engagement: A well-defined AI persona is not just a gimmick; it's a powerful tool that can dramatically increase user engagement, build a memorable brand, and make technology feel more human.

Transparency is a Feature: Being open with data and showing users that their voice contributes to a visible, collective whole is one of the most powerful features a community-focused platform can have.

What's next for Lets Roast Bolt

Our mission to make feedback better is just beginning. Next, we plan to:

Close the Feedback Loop: Introduce a "Dev Response" feature, allowing the bolt.new team to publicly attach a status or comment to feedback items (e.g., Fixed in v2.1, Good idea, we're looking into this), making the conversation truly two-way.

Launch Deeper Analytics: Enhance the public dashboard with more powerful filtering tools, allowing anyone to explore historical trends and track how sentiment for specific features changes over time.

Expand the Roast: Introduce the ability for users to roast specific documentation pages or even individual features directly, providing even more granular and actionable feedback.