Support Romanized Indian Languages in Guardrails

[InfosysResponsibleAI]

Module Name:
responsible-ai-moderationlayer

summary:
Enable risk detection ( toxicity/PII/Prompt-injection, etc) for romanized Indian Languages (e.g., Hinglish) in chatbot and app integrations.

why: Current checks miss non-standard romanized/code-mixed inputs.
Scope: Add transliteration/normalization --> run existing checks, handle common code-mix patterns.
Constraints: Sparse labeled data, high spelling variability.

[Harshil7875]

Hi, I'd be interested in contributing to this.

For romanized Indian languages (e.g., Hinglish), would the expected approach be:

1. Normalize romanized input (handle spelling variation / code-mixing)
2. Transliterate to native script
3. Run the existing moderation checks on the normalized text

Or is the goal to extend the moderation models themselves to directly handle romanized inputs?

Also noticed the issue is currently self-assigned — is it open for external contributions?

[InfosysResponsibleAI]

Hi Harshil,

Thank you for your interest in the Infosys Responsible AI Toolkit. Although the issue is currently self-assigned for coordination purposes, it is open for external contributions. We encourage contributors to align on the proposed approach before starting implementation.

We have an event scheduled on 17th March at 7:30 PM IST and we invite you to join the session to gain detailed understanding of this issue, our expectation and contribution guidelines. Join Link

[Prince777a]

Hi , ik this issue is self assigned but can i contribute to other issues here ? is this repo accepts external contributions for now ?
And also can i join the community channels of this Project ?