Doubt-to-Meme Translator: A Human-Centered Artificial Intelligence System for Transforming Complex Concepts into Memes, Stories, and Relatable Scenarios

Students frequently encounter conceptual bottlenecks in threshold STEM subjects, where abstract formalisms fail to engage entrenched intuitive misconceptions. To address this challenge, we present Doubt-to-Meme Translator, a human-centered artificial intelligence learning environment that transforms student doubts into pedagogically grounded educational memes, narrative vignettes, and everyday scenarios. Rather than treating visual humor as superficial decoration, the architecture operationalizes educational memes as structured cognitive instruments that expose misconceptions through deliberate visual-conceptual incongruity while preserving scientific rigor. The system parses natural-language input to diagnose misconceptions, immediately anchoring generation to an immutable canonical truth. Candidate explanations are synthesized via cognitive blueprints and evaluated by multi-stage critics, including an Optical Character Recognition (OCR) spatial engine preventing text collisions and a local Google EmbeddingGemma 2 model computing 768-dimensional semantic cosine alignment against the canonical definition. To verify conceptual mastery, the system administers novel counterfactual transfer probes, triggering truth-preserving explanatory pivots upon failure and scheduling spaced reviews at 1, 3, 7, 14, 30, and 60 days. The platform passes all 333 deterministic test assertions across eight test suites, cleanly classifies a 150-case development set, and achieves 81.67% accuracy, 96.77% precision, and an F1 score of 84.51% on an isolated 60-case regression benchmark, with an exploratory semantic calibration ROC AUC of 0.8413 across 29 pairs. Preliminary demonstration-based feedback involving 131 students provides qualitative impressions of the prototype. We present the work as an end-to-end system design, implementation, and technical evaluation.