Ajith Kumar V

I'm a Senior Consultant at Deloitte in India. My expertise centers on designing, building and deploying scalable machine learning solutions, with strong focus on developing robust ML pipelines for large-scale applications.

Email  /  Github  /  Linkedin  /  Medium Blogs  /  Youtube Channel  /  Scholar  / 

Interests

I am interested in generative AI and applied machine learning. Currently focusing on developing expertise in the following areas:

• Agentic AI
• MCP
• A2A
• Recommendation Systems
• Product Search
• Review Analysis
• Text Embeddings
• Customer Segmentation
• Demand Prediction
• Drift Detection
• Experiment Tracking
• Canary Deployment
• Multi-region Deployment

Notable Open Source Contributions

📦 Project: numt
Lightweight numerical utilities and transformations for Python.
📦 PyPI    💻 Source

🚀 Merged Pull Requests in Popular Repos
Adalflow: #205, #207, #226, #260, #296, #297, #289*
OpenCV: #24574, #25010, #25041
PyTorch: #95454, #95692

📌 Note:
#25010, #25041 are committed to another version of OpenCV based on code from #24574
#95454, #95692 are docs

Articles

Using AWS S3 with DVC in a GitHub CI/CD Pipeline.

Auto Start and Stop — AWS EC2 Instance with GPU.

Setting Up a Workflow with DVC, Google Cloud Storage(GCS) bucket and GitHub Actions.

Setting Up a Workflow with DVC, Google Drive and GitHub Actions.

Simple Custom Object classification with Pytorch | ONNX inference.

How to calculate GMAC and GFLOPS with python ?.

Deblurv2 GAN with custom dataset.

U-Net: Semantic segmentation with PyTorch on Custom dataset.

How to modify Cuda, GCC, python versions in colab.

Reinforcement Learning — Dots and Lines — Snake — 3/3.

Reinforcement Learning — Dots and Lines — Snake — 2/3.

Reinforcement Learning — Dots and Lines - Snake — 1/3.

Publications

Most of my research projects are related to computer vision.

The proposed model includes circle connect layers, which is the backbone of ResUNet-a architecture. The model possesses a contracting part with residual bottleneck and circle connect layers that capture context and expanding paths, with sampling layers and merging layers for a pixel-wise localization.

The proposed model includes circle connect layers, which is the backbone of ResUNet-a architecture. The model possesses a contracting part with residual bottleneck and circle connect layers that capture context and expanding paths, with sampling layers and merging layers for a pixel-wise localization.