Leads the team this role joins
Hires for this team
On the team, can refer you
From your own Gmail and LinkedIn.
About the role
This is a full-time computer vision data scientist role for engineers based anywhere in India, worked remotely, with four or more years of applied machine learning behind them. The work centers on detection, segmentation and tracking for drone and aerial imagery, and the scientist owns model quality for a multi-tenant drone analytics platform. Adapting models to each customer, building evaluation benchmarks and monitoring drift in production are central parts of the job.
What you’ll do
- Train and refine object detection and segmentation models on aerial footage, using architectures such as the YOLO family, RT-DETR, Mask R-CNN and U-Net.
- Handle small or dense objects, steep camera angles, altitude-driven scale changes, motion blur and lighting variation through tiled inference, multi-scale training and augmentation.
- Build multi-object tracking that keeps object IDs consistent across frames and flights, including camera motion compensation.
- Work with the drone data architect to convert pixel detections into geographic coordinates and measure positional error.
- Version datasets, write annotation guidelines, check label quality and split data by site and flight to prevent leakage.
- Select high-value training samples using uncertainty and diversity, and label them in CVAT or Label Studio.
- Define a repeatable fine-tuning process that keeps each tenant's data out of other tenants' models.
- Create golden test sets and report detection, segmentation and tracking metrics for each model version.
- Set regression thresholds that block promotion of any model that performs worse than production.
- Monitor input and output drift, set retraining triggers and alerts, and provide confidence calibration.
- Export models to ONNX, optimize them with TensorRT, and verify the accuracy loss.
- Deliver baseline and improved models, benchmark reports, model cards and handover artifacts to the Python ML developer.
What they’re looking for
- Four or more years in applied machine learning, including at least two years in computer vision on video or aerial imagery.
- Strong PyTorch and OpenCV skills, with experience training detection and segmentation models end to end.
- Rigorous evaluation practice, including mAP, precision-recall analysis, per-class analysis and statistical validation.
- Experience with dataset curation, annotation tooling and augmentation pipelines.
- Solid Python engineering habits: testing, packaging, reproducibility and experiment tracking with MLflow or Weights and Biases.
- Working knowledge of ONNX export and GPU training.
- A bachelor's or master's degree in computer science, data science, engineering or a related field.
Nice to have
- Tiled inference with SAHI, and SAM or foundation-model-assisted labelling.
- Multi-object tracking and video analytics.
- Drift detection tooling such as Evidently.
- TensorRT quantization and edge deployment on Jetson devices.
- Vision-language models for search or automatic labelling.
- Remote sensing or geospatial machine learning.
What’s on offer
- Listed location is anywhere in India, across multiple locations.
Skills
Questions about this role
Where is this role based?
The location is listed as anywhere in India, across multiple locations.
How many years of experience do I need?
The posting asks for four or more years in applied machine learning, with at least two years in computer vision on video or aerial imagery.
Do I need a degree?
A bachelor's or master's degree in computer science, data science, engineering or a related field is listed.
Which frameworks should I know?
PyTorch and OpenCV are required, and the stack also includes YOLO-family models, ONNX, TensorRT and MLflow or Weights and Biases.