I've been working on a new search engine for the Internet, called Scribe (https://goscri.be). It's entirely powered by AI, with LLMs and embeddings used to classify, rank, and filter results, and built with the intention to bring more quality and power user features.

Some of its features include:

- All results are evaluated using trained AI models to detect paywalled, spammy, SEO-heavy content. Results are ranked based on merit only. - Get quick, direct, human-friendly answers for questions by an AI assistant, instead of digging through long-form blog posts and documentation. - For longer, deeper discussions, continue searches by chatting with the assistant with saved and searchable conversations. - Quickly preview a page without leaving the search, to quickly evaluate or use the content. Discover related pages from across the whole web, including from unseen corners. - Personalise your results by filtering out sites you don't want to see results from. - Surface quick facts and answers from official and trusted sources without needing to dig into the page.

I'm also currently working on:

- Leave a comment and rating for a page, share and discuss with friends and family, and collectively filter or label the quality of a page or site right within the results page. - Navigate an entire site using natural language. For example, "cancel an order" on Amazon.com takes you to the cancellation page, "coffee grounds" surfaces those products within a small online shop, etc. - Within a page, quickly jump to a section using natural language, or chat with an AI assistant about the content. - Show videos from the web in the search results. For example, "max s3 buckets" highlights the exact utterance in a talk by AWS that mentions the answer. - A companion that helps you organise, track, jump between, and take notes of what you visit while deep diving, instead of having hundreds of tabs open.

This project has been an interesting technical challenge — the index is fully built from scratch (it's not backed by any existing search engine or dataset), and many parts of the stack are developed and managed entirely in-house, from the crawler, to the RocksDB cluster, to the AI classifiers.

My vision with Scribe is a web not as it is, but as it could be — a vast collective human effort of creativity and ingenuity. Where today's giants place boundaries, I dream of an alternative. I envision a digital realm where every voice, no matter how faint, finds an audience; where creativity isn't stifled by algorithms but is amplified by them. With the transformative power of AI, I'm not looking to replace the heart and soul of the web, but to celebrate and reveal it.

Scribe is still in beta — I'm still growing the index and improving the results — and I intend to open up the development and get diverse ideas from everybody. If you want to discuss ideas, feedback or issues, or want to follow our progress, connect with us on our subreddit (https://www.reddit.com/r/goscribe/) or Discord (https://discord.gg/nmUhrRfx).