Machine Learning · Research · Engineering

Vritansh Kamal

I build and study intelligent systems at the intersection of neuro-symbolic AI, deep learning, and optimization.

About Me

I am from Kullu, Himachal Pradesh, India. My work sits across machine learning, structured reasoning, and applied AI systems. I am especially interested in systems that can combine learned representations with explicit knowledge and verifiable intermediate decisions.

I hold an M.S. in Data Science from Columbia University and a B.Tech. in Computer Science from SRM Institute of Science and Technology.

Research

Primary interest

Neuro-Symbolic AI

I am interested in combining neural networks with symbolic reasoning, knowledge graphs, and ontologies to make model decisions more grounded, traceable, and structurally constrained.

Primary interest

Deep Learning Optimization

I am interested in how deep neural networks can be trained and adapted more efficiently through gradient-based optimization, regularization, and quantization, with an emphasis on smaller and more efficient models.

Industry Experience

Healthcare

CVS Health

Worked on applied data science and time-series forecasting problems for healthcare pricing and underwriting workflows.

Telecommunications

Verizon

Applied machine learning to telecom data and operational problems, working with large-scale data and predictive systems.

Enterprise AI

Deloitte · TCS

Worked across enterprise analytics, natural language processing, and applied AI systems.

Projects

Graph-Grounded Reasoning for AI Agents

Ongoing direction

Exploring how knowledge graphs and ontologies can constrain multi-step decisions in intelligent agents, particularly when systems combine retrieval, reasoning, and tool use.

Retrieval-augmented generation · Knowledge graphs · Symbolic reasoning

Digital Human Generation

Columbia University

Explored computer vision methods for digital-human reconstruction and generation, including GANs, diffusion models, NeRF, Gaussian splatting, and lip synchronization. The project culminated in a working lip-sync generation system.

PET/CT Lesion Detection

GE HealthCare

Worked on object detection for medical imaging using PET/CT data, exploring transformer-based detection and established deep-learning baselines.

Writing & Notes

Technical writing, working notes, and externally hosted articles. This section is generated from small Markdown metadata files so it can grow without redesigning the page.

All on Medium

Loading writing…