Keshaw Ranjan

I'm building AI systems across research and products, from long-context transformer methods at IIT Bombay to an AI-powered museum guide at MapIoT.ai. My work combines retrieval-augmented generation, Groq-hosted LLMs, Deepgram, PostgreSQL, Supabase, Unity, and C# to deliver conversational, interactive experiences.

In my undergraduate research, I'm investigating context extension for transformer architectures through RoPE/YaRN, attention-score redistribution, hardware optimization, retrieval-augmented memory, and hash-based context compression. I'm also working on target-based drug design, implementing diffusion models, VAEs, and GANs for molecular design and synthesis.

I built projects including AutoScientist, an autonomous researcher that searches the web and papers through a graph-structured knowledge brain; Phish-Net, an AI-backed phishing defense system; and AEGIS, an autonomous security agent for reconnaissance and controlled vulnerability testing.

I've also implemented GPU-parallel graph routing with CUDA, Wi-Fi drone control using ESP32 and Arduino, and quantum algorithms in Qiskit on IBM Quantum Cloud. I enjoy exploring the intersection of AI, security, quantum computing, interactive systems, and human-like reasoning.