Huske

I'm from the team of people who writes a lot on paper to offload ideas, tasks and thoughts to take off steam from the dozens of new things that come to mind now and then. But this isn't perfect as it tends to move the problem, and these notes tend to go to the bin and not serve their purpose.

How can "AI" or LLM come into play?

While I use AI for coding and learning about new concepts every day, I've recently realised that it can be excellent at doing entity extraction.

You can ask a Large Language Model like ChatGPT to extract anything from a text. It does it using zero-shot entity extraction. It's when a computer program can understand and recognise things, like the names of animals or people, without being taught explicitly about them. It's kind of like a magic trick that the computer can do!

What can we do with this?

How I do it

While you may use it in many ways, my idea is to use it in the same fashion I think about stuff. Something comes to my mind; I process its different concepts/characteristics to push it into a long-term or shorter thinking loop.

This is great, you would say? Well, it depends on how much stuff has to be there and how much you can handle. Other external factors may also hijack these loops and make a mess, which occurs.

How the AI may handle it

I feed it an audio note or a note. It ingests it, extracts all its concepts and understands them. It has a general notion of time and can help me understand the overall context, and can give me back things like:

• an update on the backlog of tasks
• a daily update on the progress of objectives
• an overview of thoughts and ideas
• point out things that are important to me
• and more

Is it possible?

I think so. I'm not a computer scientist, but I code.
OpenAI opened the way for people like me to be able to do it - in days.

A Minimal Viable Product

Even though this project serves more of a personal need, I still started it as an MVP to see where I was going, and the results are mind-blowing to me.

It's a two steps process:

1. I feed it a note
2. I receive a structured JSON object telling me everything it understood

I hacked a quick UI using #ArrowJS and #TailwindCSS to get started.

Let's have a look at the following example:

From feeding the following note:

Need to schedule a meeting every week with the marketing team for next Monday at 9am until 11. Invite John Smith, Jane Doe, and Sam Johnson. Probably happen in the conference room A on the 5th floor. I still have to confirm their emails and phones to include in the invite.

It managed to extract the following concepts:

• It's about a work related meeting.
• Sentiment is neutral.
• People involved are John Smith, Jane Doe and Sam Johnson.
• It's a meeting that will happen every Monday at 9am.
• It's a meeting that will take place in conference room A on the 5th floor.
• It's a meeting that will last until 11am (or 2 hours)
• The note was written in English.
• Lots of keywords.

All of these are actionable. Everything is formatted so that I can use it to create a task, a calendar event, a note, a reminder, etc.

To see more examples, check out my feed & subscribe.

What's next?

While it sounds fantastic to control everything in a chat fashion, the plan is to be able as well to navigate/search within an infinite whiteboard. This is, to me, the ultimate UI to be able to visualise and interact with all the information I have.

What do you think?
Hope you like the journey!