Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

Hi, I’m Harshavardhan Darekar, an AI Security Researcher focused on empirical evaluation of security risks in large language models and agentic systems. I specialize in prompt injection, data poisoning, and misuse scenarios, with a strong background in penetration testing and vulnerability research. I enjoy translating adversarial findings into practical safety insights to help build more reliable and secure AI systems. Currently I design and run empirical adversarial experiments for LLM-powered agent systems at Mindrift, creating automated Python workflows to explore failure modes across multiple interaction channels. I explore how adversarial inputs propagate through retrieval-augmented, tool-using, and environment-aware agents, documenting failure cases and contributing to safer, more robust AI deployments. I also collaborate with engineering teams to prioritize mitigations and improvements.…Hi, I’m Harshavardhan Darekar, an AI Security Researcher focused on empirical evaluation of security risks in large language models and agentic systems. I specialize in prompt injection, data poisoning, and misuse scenarios, with a strong background in penetration testing and vulnerability research. I enjoy translating adversarial findings into practical safety insights to help build more reliable and secure AI systems. Currently I design and run empirical adversarial experiments for LLM-powered agent systems at Mindrift, creating automated Python workflows to explore failure modes across multiple interaction channels. I explore how adversarial inputs propagate through retrieval-augmented, tool-using, and environment-aware agents, documenting failure cases and contributing to safer, more robust AI deployments. I also collaborate with engineering teams to prioritize mitigations and improvements.

Harshavardhan Darekar

AI Engineer, Data Scientist, AI Prompt Engineer, +2





Hi, I’m Harshavardhan Darekar, an AI Security Researcher focused on empirical evaluation of security risks in large language models and agentic systems. I specialize in prompt injection, data poisoning, and misuse scenarios, with a strong background in penetration testing and vulnerability research. I enjoy translating adversarial findings into practical safety insights to help build more reliable and secure AI systems. Currently I design and run empirical adversarial experiments for LLM-powered agent systems at Mindrift, creating automated Python workflows to explore failure modes across multiple interaction channels. I explore how adversarial inputs propagate through retrieval-augmented, tool-using, and environment-aware agents, documenting failure cases and contributing to safer, more robust AI deployments. I also collaborate with engineering teams to prioritize mitigations and improvements.…Hi, I’m Harshavardhan Darekar, an AI Security Researcher focused on empirical evaluation of security risks in large language models and agentic systems. I specialize in prompt injection, data poisoning, and misuse scenarios, with a strong background in penetration testing and vulnerability research. I enjoy translating adversarial findings into practical safety insights to help build more reliable and secure AI systems. Currently I design and run empirical adversarial experiments for LLM-powered agent systems at Mindrift, creating automated Python workflows to explore failure modes across multiple interaction channels. I explore how adversarial inputs propagate through retrieval-augmented, tool-using, and environment-aware agents, documenting failure cases and contributing to safer, more robust AI deployments. I also collaborate with engineering teams to prioritize mitigations and improvements.

Available to hire

Hi, I’m Harshavardhan Darekar, an AI Security Researcher focused on empirical evaluation of security risks in large language models and agentic systems. I specialize in prompt injection, data poisoning, and misuse scenarios, with a strong background in penetration testing and vulnerability research. I enjoy translating adversarial findings into practical safety insights to help build more reliable and secure AI systems.

Currently I design and run empirical adversarial experiments for LLM-powered agent systems at Mindrift, creating automated Python workflows to explore failure modes across multiple interaction channels. I explore how adversarial inputs propagate through retrieval-augmented, tool-using, and environment-aware agents, documenting failure cases and contributing to safer, more robust AI deployments. I also collaborate with engineering teams to prioritize mitigations and improvements.