hallucination-control

Hoshimiya Script / StarPolaris OS — internal multi-layer AI architecture for LLMs. Self-contained behavioral OS (Type-G Trinity).

Neurosymbolic framework that catches LLM hallucinations mid-reasoning using causal graphs and GNNs. Converts chain-of-thought outputs into Causal Reasoning Graphs, detects flawed steps with a Graph Attention Network, and auto-corrects via RAG context injection ... before errors propagate.

Evidence-grounded medical RAG system that retrieves FDA and NICE drug guidelines, generates cited answers, and safely refuses unsupported queries to minimize hallucinations.

Explicit control and observability over when an LLM should answer, hedge, or refuse — treating generation as a governed system layer, not a side effect of retrieval.

Trustworthy Agentic RAG prototype for scientific knowledge bases with multi-granularity retrieval, answer auditing, and conservative refusal.

Emergent pseudo-intimacy and emotional overflow in long-term human-AI dialogue: A case study on LLM behavior in affective computing and human-AI intimacy.

Governed AI/MCP delivery-intelligence workbench with synthetic programme data, evidence tracing, evaluation, telemetry, and visual dashboard.

A relational knowledge base of logical relations between concepts (genus-species, part-whole, cause-effect, opposition) across domain universes, designed as a grounding layer for LLMs: machine-readable definitions, validated relation grammar, and an LLM-driven filling engine.