PRISM: Progressive Resolution with Independent Section Mapping for Scalable User Profiling

Personalizing a language model over a user's long interaction history usually means either injecting the raw history into every prompt or compressing it once into a static persona. Both scale poorly: raw injection spends input tokens linear in history length, and a static persona re-generates every attribute for every query. We instead model the profile as an infinite memory: a hierarchy in which short-horizon episodic memories are compressed from narrow raw windows, and long-horizon memories are compressed from those episodic memories rather than from the raw history—so the input to any level is bounded by the level below it, and the memory absorbs unbounded history without ever feeding it all into a single prompt. We realize this memory with PRISM (Progressive Resolution with Independent Section Mapping), combining (i) incremental hierarchical compression of episodic into long-term memory, and (ii) a Map–Reduce decomposition that generates each memory level's independent semantic attribute sections in parallel, so the autoregressive generation depth is set by the longest section rather than the sum, and a broken section regenerates alone. Every efficiency result is reported in tokens, independent of any specific hardware. PRISM keeps the input tokens to refresh memory flat ($6,343$ tokens) as history grows, yielding a $\sim$63$\times$ input reduction at $10,000$ events over monolithic regeneration, and cuts autoregressive generation depth by $12.9\times$ at $K=10$ sections. Crucially, on a paired longitudinal cohort ($n=200$ accounts over 13-week histories), section decomposition significantly improves profile completeness ($\Delta =+0.384,p<0.001$) and specificity ($\Delta =+0.296,p<0.001$) under Holm-Bonferroni correction while preserving global consistency ($4.86$ vs. $4.84$) and establishing rigorous non-inferiority in deterministic episodic fact retrieval ($p<0.0001$). On downstream tasks (LaMP), PRISM matches or exceeds monolithic prediction accuracy. We release anonymized code and the LongProfile benchmark for long-horizon personalization.