The person behind the coach in your pocket
Justin Harris earned a Master’s in Atomic Physics, won Mr. Michigan, and has coached IFBB pros for 25 years — proving hard science and the iron run on the same standard. TroponinIQ is what happened when he pointed that standard at the machine doing the answering, and ended up doing original AI research most software companies never touch.
R = base + Σ (wₖ·fₖ)— the equation his coaching AI is scored on, and the one he turned into a live diagnostic
One standard, three worlds
A good physicist, a great prep coach, and a serious ML engineer all do the same thing — form a hypothesis, measure what actually happened, throw out what the data kills, and keep only what survives a check that could have failed. Justin has run that loop on particles, on physiques, and now on the retrieval system inside TroponinIQ. Same intolerance for hand-waving, three different substrates.
The conservation-law thinking behind his AI work isn’t a costume. It’s his actual training. A physicist reaching for symmetry and conserved quantities to reason about a system is doing his job — not borrowing a metaphor.
Every answer TroponinIQ gives is grounded in his own methodology — contest prep, offseason growth, peak week, nutrition, hormones — the material he’s used to bring pros to the stage. The knowledge base is his, page by page.
The builder
The engine that decides which piece of his knowledge answers your question is a reranker: it scores candidates with a weighted sum of signals. Tuning those weights is where most systems quietly go wrong — some weight-combinations are unidentifiable, so tuning them is noise. Justin built a tool, drawn from a Noether-type conservation lens, that computes those blind spots from live traffic before a single tuning run is wasted.
caught its own false positive
A signal pair looked entangled — a proper control showed it was an artifact, not real. The production map came back clean.
read-only · 0ms
Byte-identical to the live scoring path, and mutation-proven: break the math, a test fails.
arXiv-bound
Two research write-ups in progress — a theory paper on designed symmetries, and “Identifiability Audits for Additive Rerankers.”
Noether-type, before any physicist reaches for the pitchfork — the point isn’t a grand claim, it’s a tool that works on a real system. Which is exactly what you’d expect from someone who trained to distrust the impressive number until a control survives it.