Headline

Engineer ·

About

I build production AI systems from security architecture for autonomous agents to LLM pipelines that ship to real users. At NVIDIA, I work on agentic AI security: governance frameworks, adversarial evaluation, runtime policy enforcement, and observability infrastructure for production multi-agent systems. Before NVIDIA, at Samsung, I shipped LLM-driven developer tooling and multilingual NLP systems at scale with measurable outcomes: 35% latency reduction, 31% drop in customer support complaints, 60% improvement in operational efficiency. What I've built: · Autonomous agent security framework - audit instrumentation, adversarial exploitation testing, runtime policy enforcement, and PGL (Policy Grammar Language): a structured constraint system that makes LLM-driven policy verification tractable in production. · LLM-to-code pipeline - transformed design tokens and layout semantics into production UI code through iterative model-refinement workflows. · Multilingual query understanding system - contextual NLP, semantic intent mapping, dynamic dictionary expansion across languages. · Enterprise workflow orchestration - deterministic task-state transitions, live rule propagation, consistent policy execution at scale. 4 patents filed (India) spanning wearable anomaly detection, adaptive UI systems, and VR interaction . I close the loop from research to IP. I thrive in ambiguous, fast-moving environments where the output is working software in production, not decks. I've worked across hardware, ML, and software boundaries, collaborated with cross-functional teams, and translated complex technical systems into outcomes stakeholders can see. If you're deploying AI systems in real environments and need someone who owns end-to-end delivery , let's talk.

Location

Pune District

AI stack

cursor

opencode

Work experience

Engineer 2 at NVIDIA, Nov 2025 - Present · 10 mos, Pune District, Maharashtra, India, Architected an autonomous agent security framework for production multi-agent orchestration systems spanning audit instrumentation, adversarial exploitation testing, runtime policy enforcement, and automated policy synthesis. Designed PGL (Policy Grammar Language) , a structured constraint framework enabling LLM-verifiable runtime policy enforcement against prompt injection, workflow hijacking, privilege escalation, and unsafe tool execution in deployed agent systems. Built observability and evaluation infrastructure capturing inter-agent communication, tool invocation traces, execution metadata, and adversarial attack simulations , improving runtime auditability and deployment safety for production agentic systems. Engineered ML and computer-vision pipelines for video defect analysis detecting rendering artifacts, flicker, tearing, and frame inconsistencies across heterogeneous GPU hardware environments at scale. Honourable mention, NVIDIA internal Agentic AI Hackathon built a secure autonomous-agent evaluation system end-to-end under hackathon constraints and presented findings to cross-functional stakeholders.

Software Engineer at Samsung India, Jun 2022 - Nov 2025 · 3 yrs 6 mos, Noida, Uttar Pradesh, India, Engineered an LLM-driven UX-to-code pipeline transforming structured design tokens and layout semantics into production-ready UI code through iterative model-refinement workflows, reducing generation latency by 35%. Samsung Settings Search - a multilingual query-understanding system combining contextual NLP, semantic intent mapping, and dynamic dictionary expansion across languages reducing customer support complaints by 31%. Designed and shipped WorkFlow : enterprise-scale orchestration modules enabling deterministic task-state transitions, live rule propagation, and consistent policy execution across large distributed systems; improved operational efficiency by 60%. Built a real-time facial tracking and adaptive UI positioning system dynamically repositioning interface elements to prevent visual occlusion during device interaction, deployed across Samsung devices.

Intern at Incture, Jan 2022 - Apr 2022 · 4 mos, India, JSON based Dynamic Task Lifecycle Definition

Education

Bachelor of Technology, Information Technology, Indian Institute of Engineering Science and Technology (IIEST), Shibpur

Cloud Architecture using Microsoft Azure, Cloud, Udacity

Public links

github: https://github.com/variiiest