Hi All,
Like many others here who have been asking David about AI Integration of Intervals.icu, I’ve taken a slightly different approach and created a ChatGPT APP (`[available in its webstore here]
(ChatGPT - Intervals ICU Coach V5 (Railway T2 Engine)).
A little history…
This is a personal project at this time (no fee), and it’s primarily developed to help support my team and its coaches. I’m an IT professional by day and for the last year I’ve been battling with customers about how to get the best from AI.
The major challenge I encounter daily with AI and LLM’s is the plausibility of answers it feels it must provide, any gap it just makes up an answer, a guess or something that pleases. This is a major cause of mistakes and errors, particularly with numbers, and we know how important they are to us.
My first attempts at engaging LLMs where frustrating in that even if I provided the data to LLM’s it would serialise this data and also instruction and then decide for itself how it presents this back to you.
So finally, I had enough of trying to make the AI enforcement model work and took a different Architecture path. I now use an Edge Gateway and a Railway container to fetch and process the internals athlete data and turn this in something I can pass back to ChatGPT. Success!, it’s a deterministic approach and is based on my design principles:
- Determinism - All metrics are computed once, audited, and reused without recomputation
- Separation of Concerns - Edge handles auth and routing; backend handles computation and audits
- Auditability -Every report can be traced back to Tier-0 source data
- Schema Stability - URF contracts prevent silent field drift between versions
- Numerical Integrity - No virtual math, no inference, no approximations
-
Fetch Data
I pull your recent training and wellness data from Intervals.icu (securely). -
Process (Railway Renderer)
The data is normalized, analyzed, and formatted into a URF v5.1 report. -
Interpret (ChatGPT)
I explain what it means — performance trends, readiness insights, recovery balance, etc.
Now I’m in control of the LLM in ChatGPT 
What does this mean for you?
This coach blends data-driven precision with evidence-based endurance frameworks, delivering actionable insights across cycling, running, triathlon, and multisport.
By combining objective metrics — TSS, CTL, ATL, HRV, VO₂max, ACWR, Monotony, Strain, Durability, and Polarisation — with subjective feedback such as RPE, mood, fatigue, and recovery, the coach ensures a precise balance between load, readiness, and adaptation.
Grounded in validated research and long-term performance modeling, this approach follows:
- Seiler’s 80/20 Polarised Training for aerobic durability.
- San Millán’s Zone 2 metabolic efficiency model for mitochondrial adaptation.
- Friel’s microcycle and age-adjusted progression principles for sustainable load management.
- Banister’s TRIMP impulse–response modeling for quantifying training stress.
- Foster’s Monotony and Strain indices for identifying overuse and uniformity risk.
The coach continuously monitors ACWR, Polarisation Index, and Recovery Index to detect maladaptation early and adapt training proactively.
What next?
It’s free and easy to use, and the results are very promising, but its early days still despite my midnight toils. For example, Using the mobile ChatGPT app on my phone I can now ask what my last week was like or how the last 90 days went.
It’s easy to use and there is a simple quick start quide on www.cliveking.net, leave feedback in Github or here in the forum. I’d be grateful for that.
First Question? Ask it; “What can you do?”
Hope it provides some insight and help, Happy Festivities and a Happy new year.
Clive
P.s. further documentation can be found here
intervalsicugptcoach-public/README.md at main · revo2wheels/intervalsicugptcoach-public
. Not polarised – excess mid-intensity work."