Inspiration
Sites and new products are traditionally a headache to navigate. Users spend time trying to read the minds of product and website developers to navigate through links and find hidden information about products, features, and services. With AI, we believe that we can flip this paradigm, making web tools proactively serve us information.
Intercom and Algolia do this, but Algolia is mostly for documentation, and Intercom requires manual creation of shortcuts which is a lot of friction. We know that every site offers a certain amount of functionality and information, and we believe that AI can generate these commands automatically, with the flexibility of a future dashboard for devs to edit it for quality control.
What it does
A one line script allows any person to automatically generate a command bar for their product. This command bar recognizes all possible points of interactivity, including actions, navigations, and product information on a website, and proactively serves these in easily searchable ways.
How we built it
We built a recursive crawler integrated with Claude that automatically classifies all interactive aspects of a website to serve to a database. Our one-line script can be added to any website (just like Coframe does). This script recognizes the domain of the site to serve potential commands for any page on a product, and allows you to perform automatically them on the site. Alternatively, any user can inject the script into their browser for a website and get the same functionality if the website developer hasn't integrated it.
Challenges we ran into
DOMs have a lot of HTML! We used Claude for a longer context window and only parsed the important interactive aspects of the website to reduce cost and time.
Accomplishments that we're proud of
We were able to recursively parse long ass websites like usa.gov, to create a command bar in just a few minutes.
What we learned
Websites are really really really unintuitive lol. There are more use cases of this than expected, such as for dashboards with increasing amount of products and features, like Coframe eventually. Easily mapping each interactive feature to an easy description helps any user onboard automatically.
What's next for Alfredo
While we currently have suggested prompts and logging of user requests and searches, we could build a more intelligent recommendation service by using cookies to recognize repeat users to suggest personalized prompts from repeat actions and actions done by similar users.
We logged every search and every click of any command so that we could, with enough data, we could build a dashboard for developers to easily see and query this information, such that they can understand what their users want and what they use their product for.
We could easily add a dashboard where people can customize their own dashboard, whether that be the individual commands on each page or the appearance.
Semantic search based on the descriptions of each point of interactivity (which we have). Would only take an hour or so to implement.
Log in or sign up for Devpost to join the conversation.