Call for submissions: Platform Engineering meet-ups
Rootconf
Rootconf

Call for submissions: Platform Engineering meet-ups

Share your journey of building platforms that power engineering teams

All submissions
Abhijit Roy

Abhijit Roy

@Roy214

Guardrails: Keeping a General-Purpose Model in Its Lane

Submitted Aug 19, 2026

Session type: Talk (30 mins) Submission made for: Bangalore meet-ups

{Describe your session in 2 paragraphs}

A model deployed as a history assistant will still happily write you Python. Deploy an LLM for one business role and it has no idea that role exists. The usual fix is a system prompt — but a prompt is a request, not a constraint. “Only answer history questions” is an instruction the model can ignore.

This session covers enforcing that boundary properly: Sarvam-30B on vLLM with NeMo Guardrails deployed via TrustyAI on OpenShift AI / Open Data Hub, proxying every request through input rails, model, output rails. We’ll look at the config, what gets blocked and why, and the same layer masking PII and blocking Aadhaar — where it turns out Presidio ships US entities, so Aadhaar and PAN aren’t built in, and regex fills the gap by matching shape, not truth.

{Mention 1-2 takeaways from your session}

  • A prompt shapes behaviour; a guardrail enforces it. You need both instructions on the inside, a deterministic checkpoint on the outside.
  • Safety tooling carries the assumptions of where it was built.

{Which audiences is your session going to beneficial for?}

Platform / MLOps engineers, AI engineers

{Add your bio - who you are; where you work}

I am a SSME at Red Hat. My work centers on making large language models (LLMs) more efficient, scalable, and production-ready.

All submissions

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Register

Hosted by

Rootconf
Rootconf
We care about site reliability, cloud costs, security and data privacy