Build Log - 1

I've been building Farsight (name TBD) since Oct 29, 2024. In 14 days, I've been able to make a really fucking simple version of it barely work. It's cobbled together with Twilio, Make.com, OpenAI and Airtable. I also have a pretty shitty landing page/waitlist created in Framer.

A reminder to myself that I'm building for an n=1, myself.

So far, the daily logging feature seems to be working. I get the daily sms prompts, and I actually enjoy responding. I'm not sure if it's more pride and eagerness because something I built actually functions, but there is something satisfying and centering about actually checking in.

I have to think about where I am energetically and assign a number. That action forces me to think about why I might be low or high. It's a simple task, but it sends a really high signal of how I'm experiencing that day.

Here's what's broken:

• Data capture and especially the data analysis.
  • My hypothesis is that chatgpt 4o is especially good at structuring unstructured data then analyzing it, but for some reason (mostly because I'm listening to Claude), I've been building the data to be very structured. I'm wondering if that's part of the problem.
  • The analysis itself is too complicated because we're ascribing too many tags to it. I get it. We want to create some kind of system of tags to get insights over time, but I'm not sure that it's working.

Right now, here's what we're doing:

1. Checking in in the morning with two questions:
   1. What is your energy 1-10?]
      1. Tag is a number, 1-10, associated with date, check-in time (morning), and anticipated activity
   2. What are you most looking forward to today?
      1. Tag is long text, unstructured, variable length, associated with date, check-in time (morning), and energy level
2. Checking in in the evening with three questions:
   1. What is your energy 1-10?
      1. Tag is a number, 1-10, associated with date, check-in time (morning), and highlight of the day
   2. What was the highlight of today? Why?
      1. Tag is long text, unstructured, variable length, associated with date, check-in time (evening), and energy level
3. Chat then takes these raw responses and runs them through a prompt that tags and sequences the response into a required JSON structure.

The issue is that our pre-selected tagging structure doesn't (and will never) perfectly capture the responses. There will be a surprise and delight moment on Saturday, where they will get to see a snapshot of how they were feeling that week, what they looked forward to, what went well, and what didn't. There may be

What does the end user really want? (even if they don't know it)

1. Most importantly, a "forced" moment to check in with themselves. A chance to be present in an ever-moving, ever-demanding reality.
2. Knowledge that their responses are being stored and will reappear in a helpful way down the line
3. There will be a surprise and delight moment on Saturday, where they will get to see a snapshot of how they were feeling that week, what they looked forward to, what went well, and what didn't. There may be interesting correlations between energy levels and anticipated events. What are all the correlations?
   1. AM energy <> Anticipated event
   2. PM energy <> high
   3. PM energy <> low
   4. Delta between AM energy and PM energy
   5. Was the anticipated event the highlight?
   6. What causes low energy relative to lowlights?
   7. What causes high energy relative to highlights/

The database needs to capture just the number and the raw text. I need to understand what data structure suits chapped the best for this sort of thing. I think I need to ask it to tell me what kind of insights it could first capture with these inputs, then ask it how the data needs to be structured and how the prompt needs to be structured to most optimally return clean insights.

Chat then takes these raw responses and runs them through a prompt that tags and sequences the response into a required JSON structure.

Then need to update the scenario to pull from this.

Other bugs:

• I've been running into issues with Make in that for some reason it's not pulling the raw daily data into the weekly insights. I may need to start over.
• Weekly insights scenario successfully runs but with no actual analysis – data is being incorrectly pulled or formatted somewhere in the scenario

Other considerations:

• the name. I first had the name for farsight in 2017. I think I like it because the story behind it is clever once you get the product, but there are just very little good domains associated with it. I feel like I should consider more literal names like "energycheckai" or something like that?
• marketing. I'm gaining more and more clarity around the marketing, and even just testing demand. I'm considering doing a tiktok test, where I literally just go on and "facilitate" check-ins via video. Something worth trying.