ACP Intelligence™
ACP Intelligence™ began as two experiments in one: could a fitness product understand what is realistic for someone's life, recommend the right next step and learn from what actually happened—and how far could I take that idea from concept to working product using Cursor, Claude and ChatGPT alongside RAG, Supabase and Expo?
The result became an exploration of both adaptive AI experiences and the changing boundaries between design, product and engineering.
01 -- The Question
What if a fitness product didn't just give you a plan, but learned whether the plan actually worked in your life?​
02 -- From marketplace to coach
ACP started as a marketplace for finding gyms, trainers, classes and wellness experiences. I reframed the experience around the user's goal—shifting the product from Search → Discover → Book towards Goal → Plan → Move → Track → Learn → Adapt.
03 -- Designing the intelligence
I deliberately resisted a pure LLM architecture. AI reasons and synthesises, while rules handle constraints, databases provide facts and history, RAG grounds coaching knowledge, live supply determines what can actually be booked, and coaching memory captures what ACP has learned over time.
04 -- Designing for reality
Available time became a constraint rather than a preference. ACP combines goals, experience, schedule, preferences and barriers to build a realistic weekly plan instead of simply generating an ideal fitness programme
05 -- The interesting bit: learning
Personalisation wasn't enough. I wanted ACP to learn from the difference between what it recommended and what actually happened. Instead of treating a workout as simply completed or skipped, ACP captures partial completion, difficulty and barriers. Repeated evidence becomes coaching memory and can influence the next plan.
​
Explain the evidence, not the model.
​
When ACP adapts a plan, the user should understand why. Instead of asking an LLM to explain its own reasoning, I created deterministic explanations grounded in the evidence ACP actually observed.
