A workout AI app weighs cheaper models, OCR, and RAG
A workout analysis app would read from runs, rides, swims, and similar activities, then explain pace trends, plateaus, training suggestions, and even gear choices such as shoes. The data can arrive as exports from Strava or Garmin, manual logs, or of workout stats. Because some users upload , at least part of the system needs image-reading ability.
The planned pricing has cheap tiers for per-session or weekly analysis, while the top tier adds a chat feature that can answer questions about a user’s whole workout history. A likely setup is one cheap, fast model for routine batch analysis and a stronger model for the chat tier. The main cost question is whether it is wasteful to use a everywhere just because some inputs are .
A cheaper path may be to use OCR first, turn into text, and then send that text to a lower-cost model. Long workout histories may not fit well inside the forever, so the design also has to choose between RAG over the workout database and relying on very large . Haiku, 4o-mini, and are among the cost-focused model options being considered.
Key points
- The app needs to handle structured files, manual logs, and .
- Using a for every task could raise costs unnecessarily.
- OCR may turn into text that cheaper models can analyze.
- Long-term chat over months of data may need RAG instead of sending all history every time.
- A two-model setup can separate cheap batch analysis from smarter chat responses.