Shreyas Kocheri

At Rooman Technologies, I worked on assigned machine-learning modelling tasks from problem definition through evaluation. I cleaned and prepared datasets, engineered features, and trained and benchmarked supervised models.

For DeepGlobe AI, I trained a DeepLabV3+ model for land-cover segmentation, reaching 0.478 validation mIoU and 0.422 test mIoU. I deployed a CPU-compatible inference app on Streamlit Cloud, serving the model from Hugging Face Hub.

I’ve also built projects for image classification and imbalanced-data fraud detection. In TRINETRA AI, I developed a travel-planning assistant with RAG-based retrieval and a community feedback layer.

I’m the lead author of a 2025 AIJFR paper on a robotic aerial-terrestrial platform. I’m keen to apply my computer vision and predictive-analytics experience to automated visual inspection, defect and anomaly detection, and predictive quality using real manufacturing data.