$ whoami
Parshiv Kapoor
$ echo $ROLE
AI / ML Researcher & Builder
type help to explore · contact to reach me · or press ⌘K
Pre-final year B.Tech @ IIT Roorkee
I turn research into reliable, practical AI systems, working across vision-language reasoning, medical imaging, and AI agents.
About
I'm a pre-final year undergraduate at IIT Roorkee, focused on building AI systems that are robust, interpretable, and useful in the real world. My work spans medical imaging, reasoning in vision-language models through scene-graph generation, and autonomous AI agents.
I've co-authored papers at AAAI (Student Abstract Track) and ICVGIP 2025, and enjoy taking ideas from a paper to a working, dependable pipeline.
focus: [ vision-language models, AI agents & agent memory, medical imaging ]
publishing: [ AAAI 2026 (x2, Student Abstract), ICVGIP 2025 ]
building: [ Warden, FocusBoard, RecallMind ]
now: exploring the open problems in agent memoryHighlights
SeePhys Challenge @ ICML 2025
Ranked 5th globally on a large-scale vision-language physics-reasoning benchmark (2,000+ multimodal problems, 2,200+ diagrams) with a VLM pipeline that fuses schematic interpretation with natural-language reasoning.
Publications
- ICVGIP 2025: Tuberculosis bacilli detection from cytopathology images.
- AAAI 2026 (SAPP): Dual-layer Scene Graph Chain-of-Thoughts for VLM reasoning.
- AAAI 2026 (SAPP): TWIST, weakly-supervised triplet recognition in surgical video.
Amazon ML Challenge 2025
Ranked 180 of ~10,000 participants on a product price-prediction task using multimodal embeddings.