Fix: Correct Tracking Error variable in AI prompt generation to eliminate hallucination

Fix: Add support for Bytez API response schema to resolve raw JSON output

Refactor: Robust JSON extraction and grounded PM commentary for Bytez AI

Refactor: Migrate LLM integration to Bytez AI API

Refactor: Replace regex sector exclusion with LLM-based Intent Parser

Fix: Eliminate DSL hallucination - excluded+declining stocks are 'Capital Preservation', not headwinds

Fix: Robust Regex for Sentiment Guard

Fix: Add Directional Signage Logic to Attribution to prevent Negative Rally hallucination

Fix: Explicit Sector Truth Tables to prevent AI Hallucination

Fix: Realistic Benchmark construction and Passive Mode AI gate

Feat: Max Weight Constraint & NLP Logic

Initial Release: QuantScale AI Institutional Engine