Geometry-Grounded Novel-View Acoustic Synthesis
Novel-view binaural audio synthesis without Structure-from-Motion. CVPR Workshop 2026.
Professional Summary
I work on AI systems where the model is only part of the problem. That question shows up across computer vision, spatial audio, and vision-language models. Different domains, same question: what does it take for this to actually work outside the lab?
My current research is at CIVS, Purdue University, advised by Prof. Yang Ni and Prof. Chen Zhou, working on problems at the boundary of multimodal learning and real-world deployment.
A year and a half as an SRE across 11 global deployments shaped one instinct: understand how something works before building on top of it.
I am drawn to problems where the assumptions a method makes are themselves worth questioning, because that is usually where the next research question is hiding.
Education
MSc Computer Science
Purdue University
BE Information Technology
LJ Institute of Engineering and Technology
Interests
Computer Vision
Spatial Audio & Novel-View Synthesis
Vision-Language Models
Explainable AI
Industrial AI Safety
Multimodal Learning
Publications
(2026).
Visual Geometry Grounded Novel-View Acoustic Synthesis.
CVPR Workshop 2026.
(2026).
Development of Trailing Image Detection for a Melt Shop Safety Tool.
AISTech 2026.
Projects
AI Hazard Recognition System
Real-time industrial safety monitoring with blind spot handling for steel melt shop environments.
VLM Explainer: From Patches to Phrases
Causal explainability tools for vision-language model outputs using Grad-CAM, region masking, and perturbation testing.
VLM-Based Hazard Reasoning
Domain-aware VLM reasoning for industrial hazard assessment in steel manufacturing environments.
Candidate Matcher
Semantic resume ranking using MiniLM embeddings and cosine similarity, with LLM-generated match summaries.
Experience
Graduate Research Assistant
CIVS, Purdue University NorthwestResearching computer vision, spatial audio, and vision-language models.
- Geometry-grounded novel-view acoustic synthesis: feed-forward pipeline eliminating Structure-from-Motion dependency. Accepted at CVPR Workshop 2026.
- Industrial AI safety: camera-based hazard monitoring with blind spot handling. Published at AISTech 2026 and TMS/AIM 2025.
- VLM explainability: Grad-CAM token attribution and perturbation testing to verify causal influence of image regions on model outputs.
Advised by Prof. Yang Ni and Prof. Chen Zhou.
Site Reliability Engineer
Asite Solutions Pvt LtdMaintained 99% uptime SLA across 11 global production deployments (USA, UK, Canada, Australia, Hong Kong). 25+ services on Azure with strict regional isolation requirements.
- Built Python monitor for JMS consumer queues catching silent failures before user impact. Adopted as standard team tooling.
- Built zero-downtime restart automation: verified healthy instance availability before touching degraded one. Safety check first, action second.
- Full Terraform infrastructure across 11 regions. Jenkins and Ansible CI/CD pipelines. Prometheus, Grafana, ELK, Azure Monitor observability stack.
Education
MSc Computer Science
Purdue UniversityBE Information Technology
LJ Institute of Engineering and Technology
Accomplishments
Graduate Research Grant
CIVS, Purdue University ∙
January 2025
Awarded for AI Hazard Recognition research.
1st Runner-up, Research Poster Competition
Purdue University ∙
January 2025
AI Hazard Recognition System.
Full Tuition Fee Waiver Scholarship
LJ Institute of Engineering and Technology ∙
January 2023
Awarded for outstanding academic performance (2020-2023).