Headline

AI Research Engineer | Scientist-B @ CDAC | Computer Vision & Multimodal AI | Differential Privacy Research @ Georgia Tech | Foundation Models | Edge AI

About

Hi, I am a highly motivated researcher interested in working in Computer Vision and Language Modeling. I have worked on real-time Computer Vision systems to build solutions for Safe cities and Traffic Management. I have also worked on privacy-preserving Large Language Models, and RAG systems in the past. If you're interested in my work, then drop me an email - harshva27@gmail.com

Location

New Delhi

AI stack

codex

claudecode

Work experience

Scientist - B at CDACINDIA, Jan 2026 - Present · 8 mos, Delhi, India

Computer Vision Research Engineer at Metafusion, Jul 2024 - Dec 2025 · 1 yr 6 mos, India

Machine Learning Engineer at NYX, Feb 2024 - Jun 2024 · 5 mos, Bengaluru, Karnataka, India, 1. Led and managed ML applied research projects at NYX, from conceptualization to deployment, aligning with strategic business objectives, achieving a 25% improvement in project delivery timelines. 2. Developed pipelines for text-to-image, text-to-video, and image-to-image conversion, resulting in a 30% increase in content generation efficiency. 3. Fine-tuned diffusion models to improve performance and adaptability. 4. Collaborated with interdisciplinary teams, including social media, product, and design, to ensure cohesive and user-centric ML solutions.

Research Intern at Georgia Institute of Technology, Aug 2023 - Dec 2023 · 5 mos, Atlanta, Georgia, United States, 1. Co-developed Whispered Tuning, a differentially private fine-tuning pipeline to prevent training-data leakage from LLMs during inference. 2. Improved downstream task utility by nearly 8% over standard Differentially Private fine-tuning baselines while maintaining similar privacy guarantees (ϵ, δ). 3. Analyzed vulnerabilities in Large Language Models across key sectors, focusing on data privacy. 4. Evaluated OpenLLaMA-3B and OPT-2.7B on the AI4Privacy PII-masking-65k dataset.

NLP Research Intern at Bennett University, May 2023 - Sep 2023 · 5 mos, Uttar Pradesh, India, 1. Built an LLM-powered legal question-answering system using retrieval-augmented generation (LangChain, LlamaIndex, OpenAI APIs) over Indian case law and statutes. 2. Designed multi-tool agents with self-correcting query routing across search, summarization, and citation tools, improving answer relevance and citation accuracy. 3. Analysed large amounts of legal data, including case law, precedents, and legal texts, using NLP to help lawyers make more informed decisions.

Education

Bachelor of Technology - BTech, Computer Science, Bennett University

Science, DAV Public School, Aundh, Pune