Vanguardia MGSA - MGSACore Framework
Hi AIML AWG team,
At Vanguardia MGSA, we are developing MGSACore, an open-architecture framework built on top of LIT (Learning Interpretability Tool) and open foundation models (Gemma).
Our focus is twofold:
Explainability & Safety (MGSACore Shield): Visualizing attention salience token-by-token to prevent hallucinations, detect bias, and ensure research-grade data extraction from scientific literature.
Efficiency & Latency (MGSACore Optima): Pruning redundant context in system prompts and RAG retrieval chunks to optimize inference compute.
Attached is our technical One-Pager:
One_Pager_MGSACore_EN.pdf (49.1 KB)
We would love to collaborate on active OSDR AI/ML benchmarks or integrate this interpretability workflow into ongoing data projects. Looking forward to discussing this in the next monthly meeting!
Best regards,
Vanessa Montiel Ruiz | Vanguardia MGSA