I’ve been trying the (free) version of Gemini CLI to see how it can handle my rough dietary notes.
What I did:
Click for an overview of my process
- I keep a daily diary in Obsidian with very brief notes on what I ate (e.g. “2 eggs”, “beef burger with bun”, “shrimp, broccoli, sourdough bread”).
- I have a prompt that was “trained” on a few weeks of those notes, where I annotated my food diary with exact numbers taken from Cronometer, essentially “teaching” the LLM about the relationship between my rough notes and the exact dietary numbers.
- I applied that prompt to 6 months worth of the rough notes, telling it to filter out days that are obviously incorrect (e.g. total calories < 800) and then compute my macronutrient averages along with any trends.
Result
Gemini gave me this (which seems pretty accurate)
After filtering out the days with less than 800 calories, here is the analysis of your Q2 food log:
Macronutrient Averages:
- Carbohydrates: 39%
- Fat: 42%
- Protein: 19%
Click for more analysis
Trends:
- Higher Carb Intake: Compared to Q1, your carbohydrate intake has increased significantly, making it the primary energy source in Q2.
- Reduced Fat Intake: Consequently, your fat intake has decreased and is no longer the dominant macronutrient.
- Consistent Protein: Your protein intake has remained relatively stable across both quarters.
- Travel Impact: Your diet seems to be significantly influenced by travel, with periods of higher carbohydrate and fat consumption during trips.
- Post-Travel Correction: You tend to have lower-calorie days with a higher protein percentage after returning from trips, suggesting a conscious effort to rebalance your diet.
Bottom line
As with all things LLM, this is a game changer. Once the knowledge of these tools defuses some more, this dramatically lowers the bar for anyone performing self-experiments.