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:
- PII Masking Filter: A regex and NLP-driven scanner that filters and masks sensitive outputs (emails, API keys, phone numbers) before they are sent.
- Hallucination Validator: Integrates sentence-transformers or embeddings similarity comparisons to verify if the LLM output logically aligns with retrieved context vectors, blocking anomalous deviations.
- 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!
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:
I have built FAISS RAG chatbot pipelines and security telemetry systems. Please assign this issue to me. Thank you!