Skip to content

[GSSoC '26] Feature: Add LLM Response Hallucination & PII Leakage Guard Middleware #1021

Description

@Rafiaminhaj

Hello Mentor (@SdSarthak),

I would like to contribute to AegisAI by adding a guardrail middleware to prevent hallucination and PII leakage under GSSoC '26.

I propose to implement:

  1. PII Masking Filter: A regex and NLP-driven scanner that filters and masks sensitive outputs (emails, API keys, phone numbers) before they are sent.
  2. Hallucination Validator: Integrates sentence-transformers or embeddings similarity comparisons to verify if the LLM output logically aligns with retrieved context vectors, blocking anomalous deviations.
  3. Adjustable Settings Panel: React settings controls to let admins customize guard score thresholds.

I have built FAISS RAG chatbot pipelines and security telemetry systems. Please assign this issue to me. Thank you!

Metadata

Metadata

Labels

No labels
No labels

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions