Computer vision on aerial, satellite and video data, from research to production.
I build vision models and the systems that ship them: data pipelines, APIs and cloud infrastructure. 8+ years of applied ML across graduate research and industry.
Hyderabad, India · open to remote or relocation · looking for Computer Vision Engineer, Senior ML Engineer (Computer Vision) and Applied Scientist (Computer Vision) roles, especially on aerial, satellite and geospatial imagery and video.
klbm9999@gmail.com · klbm9999.github.io · github.com/klbm9999
Selected work
Production computer vision on aerial imagery
Granular AI · 2022 - Present · AI technical lead since 2024, ML engineer 2022-2023
As AI technical lead, I lead the hail and storm-damage detection models and own the data strategy, annotation standards, training, benchmarking and the path into the product.
As ML engineer, I built the roof models: roof outline detection, facet segmentation and roof-face adjacency from aerial and satellite imagery.
Inspect.Properties, end to end
I lead the conception, architecture and delivery of Inspect.Properties, Granular AI’s commercial property-intelligence product (roof analytics, weather history, storm-damage evidence, cost estimation), which reached 2,000+ registered users within 1.5 years.
I design the pipelines behind it: imagery ingestion, processing, report generation and APIs on cloud infrastructure, with the monitoring to keep customer-facing workflows reliable.
I am the primary technical contact for production issues and mentor engineers on delivery, code quality and ML workflow.
GeoEngine, a geospatial ML platform
Built, from scratch, the image acquisition, storage and curation service and the experiment-tracking service for GeoEngine, Granular AI’s geospatial ML experimentation platform with multiple datasets over a million records each, supporting annotation, training and deployment.
Research
Watch demoGAZED: Gaze-guided Cinematic Editing of Wide-Angle Monocular Video Recordings
ACM CHI 2020 · K L Bhanu Moorthy et al.
Led the algorithm design for automatically editing a single wide-angle recording into a cinematic cut, guided by where viewers look.
Paper · Demo video · Google Scholar
Writing
Detecting hail damage from aerial imageryComing soon
How the hail-damage detection work at Granular AI comes together: 15,000 damage images drawn from about 200k source images and labelled by trained expert annotators in an in-house labelling tool, and the U-Net, DINO, YOLO and SAM models behind it.
robo-litrev: evaluating RAG strategies on robotics literatureComing soon
A project to evaluate and demonstrate retrieval-augmented generation (RAG) strategies, using robotics research literature as the corpus. In active development.
Projects
Crowdsourced road-quality reporting platform for Hyderabad, built on OpenStreetMap road data. In active development.
A project to evaluate and demonstrate retrieval-augmented generation (RAG) strategies, using robotics research literature as the corpus. In active development.
Background
B.Tech in Computer Engineering, IIITDM Kancheepuram, 2013-2017.
Python · PyTorch · OpenCV · FastAPI · Flask · Docker · Kubernetes · AWS · GCS · Bash