Phenotype discovery-the data-driven identification of clinically or biologically meaningful subgroups-is fundamental to precision medicine, but conventional clustering methods require the number of clusters <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>K</mml:mi></mml:math> to be specified a priori and struggle with heterogeneous, multimodal, or longitudinal data. We introduce the GRACE Cycle (Generate hypothesis, Retrieve evidence, Align, Converge, Evaluate), a general large-language-model (LLM)-assisted framework for phenotype discovery in which a hypothesis, an LLM, and an evidence base are iteratively refined until they agree. The framework discovers <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>K</mml:mi></mml:math> as an output through Graph-of-Thought (GoT) refinement, in which an LLM reads per-cluster summary cards plus a between-cluster similarity matrix and proposes one of three moves-SPLIT, MERGE, or COMMIT-over a spectral-clustering seed. Two technical contributions enable scale: (i) a four-component prompt template integrating pairwise comparison, fairness pre-processing, and structured JSON output, and (ii) a data-feeding strategy that compresses cohorts of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msup><mml:mrow><mml:mn>10</mml:mn></mml:mrow> <mml:mrow><mml:mn>4</mml:mn></mml:mrow> </mml:msup> <mml:mo>-</mml:mo> <mml:msup><mml:mrow><mml:mn>10</mml:mn></mml:mrow> <mml:mrow><mml:mn>6</mml:mn></mml:mrow> </mml:msup> </mml:math> entities into context-budget-respecting batches via <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>k</mml:mi></mml:math> -nearest-neighbour graph sampling. We validate GRACE across three heterogeneous phenotyping problems: (1) longitudinal Long COVID subphenotyping in the NIH RECOVER cohort <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mo>(</mml:mo> <mml:mi>n</mml:mi> <mml:mo>=</mml:mo> <mml:mn>13,511</mml:mn> <mml:mo>)</mml:mo></mml:math> , where GRACE recovers three clinically distinct subphenotypes (Protected, Responder, Refractory) with bootstrap Jaccard stability <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mo>></mml:mo> <mml:mspace/> <mml:mn>0.97</mml:mn></mml:math> that are explained by a single autonomic/post-viral-fatigue axis (a 25-fold dysautonomia gradient, dysautonomia adjusted <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>O</mml:mi> <mml:mi>R</mml:mi> <mml:mo>=</mml:mo> <mml:mn>13.4</mml:mn></mml:math> ) and an accompanying collapse of wearable-measured physical activity; (2) motor subphenotyping of Parkinson's disease from foot-sensor gait wearables (PhysioNet gaitpdb, <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>n</mml:mi> <mml:mo>=</mml:mo> <mml:mn>93</mml:mn></mml:math> ), where GRACE discovers two gait subtypes without specifying <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>K</mml:mi></mml:math> that are externally validated against withheld Timed-Up-and-Go <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mo>(</mml:mo> <mml:mi>p</mml:mi> <mml:mo>=</mml:mo> <mml:mn>0.002</mml:mn> <mml:mo>)</mml:mo></mml:math> , Hoehn-Yahr stage <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mo>(</mml:mo> <mml:mi>p</mml:mi> <mml:mo>=</mml:mo> <mml:mn>0.03</mml:mn> <mml:mo>)</mml:mo></mml:math> , and age; and (3) additional open wearable chronic-disease cohorts processed with the identical pipeline. Across domains, GRACE converges without prior knowledge of <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>K</mml:mi></mml:math> , demonstrating that LLM-guided iterative reasoning offers a domain-agnostic alternative to conventional clustering when the number of phenotypes is unknown.
