Idea

Build a layer on top of an AI (could be any and all of them) that will help a user to achieve a physical activity related goal in 1 year. It creates a plan tailored to the user’s goal, physical characteristics and limitations (injuries, pain, weight, height) and available equipment. The improvement over regular AI conversations is in the way it stores the data. Every message sent to the AI will have full context of the user’s previous interactions and training data and when this data becomes too much for a reasonable context window, it will compress it in a way that allows AI to retain context by giving up some details. From my tests, the context window lasts a long time before being compressed and when it is, its not noticeable unless you ask for a very specific piece of information like “How many reps of pull ups did I do exactly 1 year ago”.

Because of the nature of how we communicate with AI (chat), a good option to avoid FE work and reach more people is to develop this as a Telegram Bot. The API is powerful and responsive and Telegram is growing. I have also used it before and have a bot already that can be used for a prototype.

Tech Stack

• Golang with custom built openAI integration and metrics
• Managed Redis (Upstash) used as persistent data storage.
• Telegram API

Example:

Metrics Information, only available to the admin.

Example of the daily message with that day’s workout plan.

• Metrics can be saved by the admin or automatically every 5 min. Can be reset by the admin.
• User information can be removed by reseting the progress. Just need to say you want to reset, confirm and the data is deleted. The interaction can be restarted.
• The plan can be adapted according to progress.

Bot V0.1 (back end disabled)

TODO

22/03/2025

• Keep track of dates where the user does things. Days should not be incremented 1 by 1 based on the user using the app, but on the current date and previous dates. Maybe date should even be a part of dates.