Pulse Analysis

AI powered Monitoring and Analysis of public sentiment of your product.

Comment

Inspiration

We have both built B2C products on small teams (startups with fewer than 10 people). Analytics and responsiveness to user pain points and sentiment are vital to launching successful products.

What it does

This tool, when given a product, will:

  1. Populate a basic description of the product and links to publicly available changelogs
  2. Scan the internet for publicly available feedback from users
  3. Aggregate that feedback into time-range-based summaries
  4. Display a timeline showing product releases and feedback analysis overlaid on it

How we built it

This is a Next.js app. It consists of an agent (using the Anthropic SDK), powered by Nimble for web scraping and ClickHouse for data storage. The agent constructs analysis in four stages:

  1. Populates general information about the product
  2. Generates a history of releases
  3. Scrapes the web for real feedback from users
  4. Analyzes that feedback and generates summaries

This is then displayed in a scrollable timeline that shows releases and analysis side by side.

Challenges we ran into

The biggest issue was optimization. Web scraping takes time. Ideally these things could run as background tasks overnight and backfill data over time, but testing in real time was challenging. We seeded test data so that the visualization could be worked on even as we were solving data population issues.

Accomplishments that we're proud of

Figuring out how to get the agent to break down processing into shorter timelines (week by week), in order to demonstrate the product with real user data.

What we learned

ClickHouse was super easy to use. It was my first time completely vibe-coding an app like this — as a full-stack engineer, I generally use these tools in a more limited and directed way, so that was really fun!

What's next for Pulse

Ideally we would generate more helpful breakdowns of this data, specifically showing real-time user concerns and issues broken down by platform (mobile, desktop, web, etc.). We'd also like to pull in logs, issues, and internal timelines from a team to help small teams respond to user concerns faster.

Built With

Share this project:

Updates