Senior Applied ML Engineer

Qrata; Bengaluru; On-site; INR 3,600,000 - 6,000,000; Full time

Job summary

Qrata seeks an experienced ML engineer to build computer-vision and multimodal systems that understand real-world video. You will own the full model lifecycle from problem framing to production launch and post-launch improvements, focusing on reliable production behavior rather than just benchmarks.

You will work on datasets, annotations, baselines, and evaluation systems, with a strong emphasis on privacy, calibration, and real-world performance across diverse conditions.

Qualifications

• Strong Python and PyTorch experience.
• Experience shipping computer-vision models into production.
• Strong foundation in object detection and segmentation, OCR, multi-object tracking, action recognition, temporal localization, and video-language models.
• Strong understanding of precision/recall trade-offs, calibration and threshold selection, dataset leakage, label quality, distribution shift, and stratified evaluation.
• Experience building datasets and evaluation systems for messy real-world inputs.
• Ability to independently own ambiguous ML problems from framing through production.
• Strong software-engineering fundamentals.
• Ability to explain complex model behavior to product and operations teams.

Responsibilities

• Build representative training and evaluation datasets from field footage.
• Design annotation guidelines, sampling strategies, hard-negative mining, and active-learning workflows.
• Define model metrics connected to product outcomes: privacy-critical false negatives, precision and recall, confidence calibration, human-review burden, and performance across operating conditions.
• Establish strong baselines, experiment tracking, model cards, and launch criteria.
• Diagnose failures frame by frame and convert patterns into data, modeling, or product improvements.
• Use production failures and reviewer feedback to improve datasets and models.
• Work with the ML Evaluation engineer to define quality standards and regression tests.
• Work with backend and platform engineers to package, deploy, monitor, and roll back models.
• Communicate model limitations, uncertainty, and trade-offs clearly.

Skills

Python

PyTorch

Computer vision

Production ML

Object detection

Segmentation

OCR

Multi-object tracking

Action recognition

Video understanding

Dataset construction

Evaluation systems

Software engineering

Model explainability

Tools

Python

PyTorch

OpenCV

Job description

You will build computer-vision and multimodal systems that understand real-world video.

You will own the full model lifecycle: problem definition, data strategy, experimentation, evaluation,

production launch, and post-launch improvement. Success means reliable production behavior — not benchmark performance or impressive demos alone.

Must Have - Build Computer-Vision and Multimodal systems

What You'll Work On:

• Face, person, object, and sensitive-text detection
• Object and person tracking
• Temporal event and action recognition
• Video quality assessment
• Evidence extraction and video summarization
• Multimodal video understanding

What You'll Do:

• Build representative training and evaluation datasets from field footage
• Design annotation guidelines, sampling strategies, hard-negative mining, and active-learning workflows
• Define model metrics connected to product outcomes: privacy-critical false negatives, precision
• and recall, confidence calibration, human-review burden, and performance across operating
• conditions
• Establish strong baselines, experiment tracking, model cards, and launch criteria
• Diagnose failures frame by frame and convert patterns into data, modeling, or product improvements
• Use production failures and reviewer feedback to improve datasets and models
• Work with the ML Evaluation engineer to define quality standards and regression tests
• Work with backend and platform engineers to package, deploy, monitor, and roll back models
• Communicate model limitations, uncertainty, and trade-offs clearly

What We're Looking For

• Strong Python and PyTorch experience
• Experience shipping computer-vision models into production
• Strong foundation in several areas: object detection and segmentation, OCR, multi-object tracking, action recognition, temporal localization, and video or vision-language models
• Strong understanding of precision/recall trade-offs, calibration and threshold selection, dataset leakage, label quality, distribution shift, and stratified evaluation
• Experience building datasets and evaluation systems for messy real-world inputs
• Ability to independently own ambiguous ML problems from framing through production
• Strong software-engineering fundamentals
• Ability to explain complex model behavior to product and operations teams