Open to opportunities

Neha S

@nehas2

Message

AI safety researcher building deployable, multilingual LLM jailbreak defenses.

India

Message

What I'm looking for

I’m looking for roles where I can build deployable LLM safety and multilingual evaluation systems—combining red-teaming, adversarial robustness, and real-world monitoring—so models stay aligned in sensitive, low-resource settings.

I’m an M.Tech researcher at DRDO-DIAT focused on LLM safety, adversarial robustness, and multilingual AI alignment. I designed and deployed SAFE-NEST, a six-stage, model-agnostic safety pipeline that cut jailbreak success rates from 23.3% to 3.3% across six open-source LLMs, and I led evaluation work across 5+ languages at DRDO DYSL-AI—spotting a 15-percentage-point safety gap between high- and low-resource languages.

I build safety systems that don’t just measure risk—they reduce it in practice. I curated sensitive-domain evaluation datasets that improved refusal consistency by 20%, and I translated safety insights into guardrail refinements for regional-language deployments. From research to production, I’ve shipped a FastAPI service with audit logging and a React monitoring dashboard, demonstrating deployability on consumer-grade hardware, alongside complementary experience in computer-vision and full-stack ML projects.

Experience

Work history, roles, and key accomplishments

AI Research Intern

DRDO DYSL-AI

Aug 2025 - Apr 2026 (8 months)

Architected a multilingual LLM safety evaluation benchmark across 5+ languages, uncovering a 15-percentage-point safety gap between high- and low-resource models. Curated a sensitive-domain evaluation dataset that improved refusal consistency by 20%, informing safety-guardrail refinements for regional-language deployments.

Multilingual Evaluation Jailbreak Detection Cross Lingual Alignment Analysis Sensitive Domain Probing GuardRails

Research Intern

IIITDM Kancheepuram

May 2023 - Jun 2023 (1 month)

Optimized intrusion detection systems for high-performance computing by applying vulnerability mapping, reducing false-positive alerts by 20%. Authored a bug-bounty methodology report identifying 3 critical infrastructure gaps.

Intrusion Detection Systems High Performance Computing (HPC)Bug Bounty

Education

Degrees, certifications, and relevant coursework

Defence Institute of Advanced Technology (DRDO-DIAT)

Master of Technology (M.Tech), Data Science

2024 - 2026

Grade: CGPA: 8.0/10

Activities and societies: M.Tech thesis: SAFE-NEST (six-stage LLM safety pipeline) in collaboration with DRDO DYSL-AI; multilingual evaluation benchmarking across 5+ languages.

M.Tech researcher focusing on LLM safety, adversarial robustness, and multilingual AI alignment. Designed and deployed SAFE-NEST, a six-stage, model-agnostic safety pipeline that reduced jailbreak success rates from 23.3% to 3.3% across multiple open-source LLMs.

AMC Engineering College (VTU)

Bachelor of Engineering (B.E.), Artificial Intelligence and Machine Learning

2020 - 2024

Grade: CGPA: 8.4/10

Activities and societies: Final year project (AINA): virtual try-on web app using OpenCV/OpenPose and GAN-based image synthesis plus a recommendation engine.

B.E. in Artificial Intelligence and Machine Learning with a focus on neural networks and computer vision. Built a full-stack virtual try-on and recommendation platform (AINA) combining pose tracking, GAN-based clothing synthesis, and visual recommendations.