OmniSyntax — Hybrid Syntax Detection

Rule-based linters miss fuzzy errors; pure ML is opaque and fails silently. OmniSyntax runs deterministic AST checks first, falls back to a trained classifier for the rest, and degrades gracefully to rules-only when the model can’t load.

• 01Deterministic static analysis + AST parsing as the first pass.
• 02TF-IDF features → gradient boosting classifier for cases the rules miss.
• 03Metadata-aware compatibility control with an explicit "degraded mode" when the ML artifact is missing or incompatible.
• 04Auto-fix suggestions with tutor-style explanations.

1. Source → AST / rule layer
2. TF-IDF + Gradient Boosting classifier
3. Compatibility / degradation guard
4. FastAPI · Streamlit · CLI entry points

• Graceful degradation: the service stays useful (rules-only) if the ML model is absent or incompatible.
• Validating across five language grammars with 193 tests plus adversarial cases.

• Per-language: JS 97.1% · C 94.4% · Java 94.2% · Py 93.6% · C++ 91.7%
• 0.0% false-positive rate on clean code
• Graceful "degraded mode" fallback to rules-only
• Auto-fix suggestions with explanations
• FastAPI REST + Streamlit UI + CLI