Machine Learning & Computer Vision Engineer
Pune, IndiaFull-time or contract
We are looking for a hands-on Machine Learning and Computer Vision Engineer who can take open-ended technical problems, identify suitable approaches, and independently run experiments from data collection to final recommendation.
This role is ideal for someone who understands ML architecture, enjoys research-driven development, and can decide when to use deep learning, classical computer vision, geometric methods, or a hybrid approach.
What you'll work on
- Image and video understanding.
- Object detection, segmentation, keypoint detection, pose estimation, and tracking.
- Temporal analysis of actions and multi-step processes.
- Multimodal learning using visual, sensor, and time-series data.
- Failure detection, confidence estimation, and automated verification.
- Data collection, annotation, training, and evaluation pipelines.
- Integration of trained models into larger software systems.
You will review relevant research, establish baselines, compare model families, analyse failures, and recommend the most practical technical path.
What we're looking for
- Strong Python and PyTorch experience.
- Good understanding of modern computer vision and deep-learning architectures.
- Experience working with image or video datasets.
- Knowledge of detection, segmentation, tracking, pose estimation, or action recognition.
- Ability to design controlled experiments and define meaningful evaluation metrics.
- Experience with Linux and GPU-based model training.
- Ability to understand research papers and convert them into working experiments.
- Good understanding of data collection, labelling, training, validation, and model integration.
Good to have
- Detection and segmentation: YOLO, RT-DETR, Mask R-CNN.
- Pose estimation and tracking: ViTPose, Keypoint R-CNN, ByteTrack.
- Video and temporal understanding: VideoMAE, Video Swin Transformer, MS-TCN.
- Learning-based control: Behaviour Cloning, Action Chunking with Transformers, Diffusion Policy.
Experience with multimodal learning, synthetic data, domain adaptation, uncertainty estimation, or robotics-related applications is an added advantage.
Who will fit this role
You should be comfortable owning an experimentation track independently, from understanding the problem and identifying the required data to training models, comparing approaches, and presenting a clear recommendation.
We are looking for someone who can help decide what should be built, which approaches should be tested, and how to prove that the selected solution works.
Open to full time, or a long-term contract of 3 to 6 months.
If this is your seat
Send the application. We read every one, and we write back when it's a fit.