research

Two papers. Read both.

The full architecture of Serena is documented across two academic papers. The dissertation paper grounds the symbolic core. The lifespan paper extends it with non-semantic affect and developmental priors.

A Symbolic Affective Architecture for White-Box Emotional AI

Presents a fully symbolic affective AI architecture in which a 12-dimensional emotion vector is derived from PMI lexicon weights, Kneser-Ney smoothed n-gram conditionals, and ARPAbet phonemic mappings. Short-term memory is implemented over LMDB-backed adaptive resonance structures; long-term memory uses minimal-perfect hashing for constant-time symbolic recall. OCEAN personality parameters modulate response shape via a deterministic template generation stage. Every emotional output is reproducible from the trace, and no neural network or statistical language model is involved in the affective inference path.

Harvell, B. (2025). A Symbolic Affective Architecture for White-Box Emotional AI. BSc Dissertation, Bournemouth University.

Lifespan-Aware Affective Computing on a 7D Non-Semantic Vector

Extends the symbolic affective core with a 7D affective vector (VAL, ARO, DOM, PRD, NOV, COM, SRF) derived from non-semantic structural features of input — independent of the lexicon. Introduces a 6D DNA trait vector (STAB, OPEN, AGREE, ASSERT, SENSE, LEARN) and a developmental lifespan window from 0–25y, with newborn blending, homeostatic integration, and emotional plasticity. Together, the lifespan model produces age-conditioned affect and stable agent identity.

Harvell, B. (2025). Lifespan-Aware Affective Computing on a 7D Non-Semantic Vector. Beyond the Box / GlassMind Division. Working paper.