I built AI-powered web research before ChatGPT could search the web
API that searches Google, scrapes pages in parallel, and synthesizes cited research reports with AI.
I was experimenting with content automation and hit a wall — AI could write, but it couldn't research. Tools like Perplexity were just emerging, ChatGPT couldn't browse the web yet, and there wasn't a good way to programmatically feed real-time information into a prompt. So I built the research layer myself.
One endpoint takes a query and a research objective, searches Google, scrapes the top results in parallel, converts the HTML to markdown to keep token costs down, and synthesizes everything into a cited report. Configurable sources, Vancouver-style citations, cost tracking built in. I also used this as an excuse to build in Deno and explore API-as-a-product as a format.
Search-scrape-synthesize is everywhere now. At the time, I was stitching it together from raw APIs because none of the turnkey options existed yet. I used it to power my own content workflows before moving on to other things.