Veovo

• Built and operationalized Kubeflow pipelines to support end-to-end ML experimentation at scale, including automated data extraction from PostgreSQL and S3 data lake, feature engineering, model training, hyperparameter tuning, and rigorous validation with baseline comparisons and deployment of regression and time series models.
• Built and maintained Bamboo-based CI/CD pipelines for ML and software services, including automated training validation, unit/integration testing, artifact versioning, and deployment to Kubernetes environments.
• Designed and deployed cloud-native microservices on Kubernetes, integrating REST APIs and Kafka-based messaging to support scalable data processing, model serving, and distributed ML workflows.
• Collaborated with distributed cross-functional teams across engineering and product functions to define requirements and deliver production solutions.
• Designed and implemented optimization models, translating business requirements into mathematical constraints and objective functions to generate scalable, high-quality scheduling solutions.
• Tools: Python, PyTorch, Docker, Kubernetes, Kubeflow, S3

OTE SA

• Contributed to the architecture and design of large-scale distributed network infrastructure, including a $3.4M IP/MPLS VPN system (NETVIS) connecting 84 global sites.
• Applied technical feasibility analysis, system design principles, and cross-functional collaboration to define scalable, reliable infrastructure under real-world constraints.
• Tools: IP/MPLS, VPN, Network Architecture

Published MSc research in a peer-reviewed conference. Designed and evaluated a cascaded 3D Autoencoder–Visual Language Model framework for abnormal scene detection and mitigation of out-of-distribution failures in surveillance video. Integrated and fine-tuned Video-LLaMA using PEFT/QLoRA, performed large-scale distributed training on SLURM-managed GPU clusters, and benchmarked against baseline methods.

• Tools: PyTorch, Video-LLaMA, PEFT, QLoRA, SLURM

Created synthetic low-resolution and low-light images using Stable Diffusion. Trained on a multi-GPU cluster and evaluated on a super-resolution framework — results showed significant enhancement in light & image quality.

• Tools: Stable Diffusion, GANs, PyTorch, multi-GPU training

Built a multimodal Visual Inspection Model providing localization results in a 3×3 grid with textual explanation using an LLM fine-tuned with LoRA (matrix decomposition for efficient weight training). Investigated in the Industrial Anomaly Detection domain on MVTec and VisA datasets.

• Tools: PyTorch, LoRA, Vision-Language Models, MVTec, VisA

Compressed audio signals using Implicit Neural Representations. Optimized through quantization and pruning techniques. Also engaged with research on meta-learning for faster convergence across signal types.

• Tools: PyTorch, INR, quantization, pruning, meta-learning

Languages & Databases

Python

Java

PostgreSQL

Frameworks & Libraries

PyTorch

TensorFlow

Numpy

Pandas

OpenCV

Scikit-learn

Tools & Platforms

Git

AWS

Shell

Kubernetes

Docker

Kafka