Frosty Byte

Code: https://github.com/kerorokerori/Frostybyte
Project Overview
Our project is a GenAI-powered web intelligence tool designed to extract structured insights from unstructured webpages. It tackles the ASB challenge by automating data collection and understanding from online sources, turning messy web content into usable, structured formats for business or compliance intelligence.

Core Functionality
Crawling & Scraping:
We use Crawl4AI, a tool that lets users input any URL. It automatically fetches the page’s HTML structure.

HTML Cleaning & Text Extraction:
We clean the raw HTML by removing irrelevant elements (e.g., scripts, ads) and extract meaningful text content.

GenAI-Powered Structuring:
The cleaned text is passed to a Large Language Model (LLM) (Gemini) using a carefully designed prompt. The LLM analyzes and converts the unstructured content into a structured format (e.g., JSON), based on our schema or specific business context.

Our Prompt:
“””
Extract campaign data from the text below. Return structured JSON with two arrays: “LiveCampaigns” and “PreLiveCampaigns”.

For each Live Campaign, extract:

• title
• description
• categories (array)
• amount_raised (string or null)
• days_left (integer or null)
• valuation (string)
• min_target (string)
• investors (array or null)

For each Pre-Live Campaign, extract:

• title
• description
• categories (array)
• valuation (string)
• min_target (string)

Return null if a field is missing
“””