Job Tile : Data Scientist — Video Analytics
Experience: 1- 3 years
Location: Kondapur HYD( Hybrid)
ROLE SUMMARY
Build and deploy the computer vision models that power the live app— from PPE and cross-contamination detection to wastage estimation and environmental-hazard alerts. You will own the full model lifecycle: data modeling, annotation pipeline design, training, evaluation, and live deployment, working closely with the Lead Architect and the interns.
KEY RESPONSIBILITIES
- Data modeling — design data models and labeling taxonomies for hygiene, wastage, and environmental-safety detection tasks (PPE compliance, spill/pest detection, cross-contamination, portion variance, veg/non-veg segregation).
- Annotation pipeline — build and maintain annotation workflows and define QA standards for the Data Annotation Intern.
- Model training — train, fine-tune, and evaluate object detection and action recognition models (e.g., YOLO-family, pose estimation) on kitchen video data.
- Edge optimization — quantize and prune models to meet sub-2-second inference latency targets on edge hardware.
- Live deployment — support integration of trained models with the edge-inference and orchestration pipeline in production.
- Model monitoring — track deployed models for drift and accuracy degradation, and drive retraining strategies.
- Scoring engine — collaborate with the Lead Architect on the weighted Hygiene Trust Score algorithm.
- Governance — document model performance, limitations, and bias-audit results for the Ethical AI review process.
REQUIRED SKILLS & QUALIFICATIONS
- 1–3 years of experience in video analytics — data modelling, annotation, model training, and deployment.
- Proficiency in Python and a deep learning framework (PyTorch or TensorFlow).
- Hands-on experience with object detection frameworks and pipelines (YOLO, Detectron2, or similar).
- Familiarity with annotation tools (CVAT, Labelbox, or similar) and annotation QA processes.
- Understanding of model evaluation metrics (mAP, precision/recall) and edge-deployment constraints.
- Comfortable working with real-world, noisy commercial-kitchen footage — variable lighting, occlusion, motion blur.
PREFERRED / NICE-TO-HAVE
- Exposure to action recognition or pose-estimation models.
- Experience with MLOps tooling such as MLflow or Weights & Biases.
- Prior work in food safety, retail, or industrial-safety computer vision.
Pay: ₹500,000.00 - ₹800,000.00 per year
Experience:
- Video Analytics: 1 year (Preferred)
- Data science: 1 year (Preferred)
Location:
- Hyderabad, Telangana (Hyderabad District) (Preferred)
Work Location: In person
