Applied AI research with an engineering bias: systems that can be evaluated, defended, and shipped — from autonomous ML pipelines to hospital knowledge bases.
Context-aware problem canonicalization, multi-gate ethics/feasibility checks, semantic dataset discovery, AutoML (RF / LightGBM / XGBoost), and automated code generation. Evaluated on 33 problems: Acc/R² 0.690, F1 0.746, ~5 min E2E, 85% unsafe-task filtering. Research paper in prep with Garv Bansal & Vaibhav Raj.
Worked with hospital IT on digital pen / dot-paper digitization of handwritten records, vendor workflows, and AI/ML training for handwriting. Built a RAG chatbot over hospital docs and researched private open-source LLM deployment for healthcare privacy.
Kai simulates emotions, long-term memory, personality, and regulation — designed to feel less like a chatbot and more like a digital personality. Explores human-centric applied AI research.
Defense-in-depth: ML detector on 100k+ prompts, rule filters, benign whitelisting, confidence escalation. ~99–100% recall, <2% FN, >50% FP reduction. Hybrid verdict engine for agent/tool safety.
Redesigning Indian agriculture with AI-powered precision: gene-editing insights + real-time digital twin ecosystems for sustainable farming. Built for ANNAM.AI Hackathon 2025 (4th place) at IIT Ropar — CoE of the Ministry of Education.
See also AI projects and the full project index.