Thái Anh Đức

May 10, 2026

Evolve my AI Coach

Two weeks ago, I built a CLI application that allows me to register my training, very customized to me. It then connects to my Strava and acts as an AI Running Coach.

The command to log a strength training session
image.png


It has a fixed "weight". Which is not accurate for my training. Simple and Sinister program has different weights for Swing and Turkish Get Up. Train with progressive overload principle. The "notes" field is for those detail.

It does not look good. Back to the design board with Claude Code.
There is a feature in the app: log-strength. I train following StrongFirst programs, for example Simple and Sinister, usually perfom 10 reps swing with 10 sets, then 1 rep Turkish Get Up with 10 sets. Use Progressive Overload approach. Ex: train with 24kg swing, then after a few weeks, perfom 2 sets with 32kg, same for Get Up. How do we model that? Suggest some approaches.

Claude Code:
image.png


The suggested data model looks good. It fits well with my training session of various kinds. For example, in AXE program, and I usually squat before the main session.

What the interface looks like? How do I record my session?
how does the API look like? example of how I will log my session

Here it is
image.png


All clean and clear. Ready to implement.

Good Sunday morning.

About Thái Anh Đức

Software architect from Việt Nam. Partner at https://ritvn.com/.  Train with Kettlebells. Run with minimal shoes and sandals.