Hey there, I'm Ramesh
Senior Applied Machine Learning Engineer at Planet Labs
I have more than 7 years of experience building computer vision systems. These are programs that learn to understand images. I use them on satellite images and map data, from deep forests to dense cities. My work sits where AI, environmental monitoring, and large-scale software meet.
I like to turn research ideas into real products. I believe machine learning can do more than save time. It can help the planet.
What I'm Working On
At Planet Labs, I lead the work on machine learning (ML) pipelines. A pipeline is a chain of automated steps that turns raw data into results. Ours analyze satellite images at very large scale. I focus on four things right now:
- Building high-performance models that find objects in images (detection) and label every pixel (segmentation), for infrastructure and land use
- Designing workflows that train models from few or imperfect labels (weak supervision), so we can improve models quickly
- Combining different kinds of data in one model: images over time, different light wavelengths (spectral), and radar (SAR), which can see through clouds
- Making pipelines cover the whole globe, with little delay between a new image and a result
I also work on making models easier to explain (interpretability). I work with other teams to deliver tools that scientists, governments, and humanitarian groups use.
A Bit of Background
I have a Master's degree in Electrical Engineering from Rochester Institute of Technology. I focused on computer vision and pattern recognition. Before Planet, I worked at Siemens Corporate Research. There I built new methods for anomaly detection (finding things that look unusual) and segmentation.
I have published at top conferences, including ICML (the International Conference on Machine Learning). I also hold a patent in weakly supervised learning.
Let's Connect
Do you want to work together, learn about my work, or talk about geospatial AI (AI for maps and satellite data)? I would love to hear from you. Visit the contact page or use the links in the footer.