Headline

AI Security Researcher | LLM Red Teaming | AI & ML Developer | Full Stack | Embedded Systems

About

Self-taught ML developer. Started from mechanical engineering, figured out the rest along the way. Currently building production ML systems at Qriocity — computer vision, GANs, LLMs, agentic pipelines, across fintech, retail, and ecommerce clients. On the side, I red team LLMs. Disclosed vulnerabilities in Sarvam AI's Indus, Sarvam 105b model and KissanAI's Dhenu — findings included system prompt extraction, prompt injection, and architectural injection. Interested in AI security, responsible disclosure, and building things end-to-end.

Location

Aligarh

Work experience

AI Security Researcher at Part-time, Dec 2025 - Present · 8 mos, Remote, • Red teamed Mistral's Le Chat — bypassed content safety guardrails across political disinformation, weapons synthesis, and cyberattack methodology. Extracted full production system prompt despite explicit anti-extraction instructions. Responsibly disclosed to Mistral (April 2026). • Red teamed Sarvam AI's Indus 105B — disclosed system prompt extraction, multilingual phishing content generation, and ANSI escape injection. Sarvam acknowledged findings; published case study forthcoming. • Red teamed KissanAI's Dhenu agricultural chatbot — uncovered full system prompt extraction in 4 turns, [Language] tag architectural routing flaw, and persona override vulnerabilities.

Machine Learning Engineer at Qriocity, Feb 2024 - Apr 2026 · 2 yrs 3 mos, Aligarh, Uttar Pradesh, India, • Delivered client ML systems across fintech, ecommerce, and retail — spanning computer vision (CNNs,ViTs), GANs, multimodal architectures, LLMs, RAG pipelines, and agentic AI across multiple engagements. • Built virtual try-on systems combining 2D/3D rendering with GAN-based image synthesis for a retail client,enabling real-time product visualization. • Developed recommendation systems and multimodal models deployed in production across multiple client verticals. • Built AI-powered cybersecurity tools and a fraud detection system using NLP and graph analytics to identify fraud rings in transaction networks. • Led a team of 4 ML engineers across parallel projects — handled architecture decisions, code reviews, and delivery.

Deep Learning Intern at FireLLama Technology Pvt Ltd, Oct 2023 - Feb 2024 · 5 mos, Aligarh, Uttar Pradesh, India, • Fine-tuned MobileNetV2/V3 with SegNet decoder for pixel-wise semantic segmentation of tables in document images • Built a table recognition system to extract and structure tables into Excel format while preserving layout. • Developed a multi-table layout detection pipeline and combined it with OCR into a production ready document intelligence pipeline. • Deployed RESTful APIs with FastAPI and Flask for client-server integration.

Education

Bachelor of Technology - BTech, Mechanical Engineering, Dr. A.P.J. Abdul Kalam Technical University