AI New Mirror Engine (github.com)

Allen Moysey-Groat announced the Mirror Engine, a modular, locally runnable “semantic reflection” system that turns journal entries, dialogue transcripts and subjective reports into structured cognitive artifacts to support introspection and self-modeling. The pipeline extracts concepts as graph nodes, tracks frequency, sentiment polarity and adjacency, detects semantic inversions and recursive loops, and plots entry-by-entry emotional tone. Outputs include interactive graph JSON (D3.js/Cytoscape compatible), contradiction reports with polarity-swing metrics, loop indices with recurrence metadata, time-series CSV/line graphs, and spoken playback (.mp3/.wav) via offline TTS (pyttsx3 or configurable gTTS). The stack leverages TextBlob/VADER/custom sentiment rules, NetworkX/Neo4j for graphing, and a custom swing-detection algorithm; it can also integrate LLM summarization layers and a GUI overlay for belief graphs and drift maps. This matters to the AI/ML community as a privacy-preserving tool for affective and cognitive analytics: it demonstrates how lightweight NLP, graph tooling and local TTS can enable real-time introspective analytics without cloud dependency, useful for researchers, therapists, cognitive toolmakers and creative systems. Technically, Mirror Engine offers reusable structured outputs for downstream modeling (belief networks, contradiction detection, recurrence indices) and a testbed for recursive affective modeling in autonomous agents, while explicitly prioritizing user control and non-prescriptive reflection rather than behavior optimization.