Senior Applied AI/ML Engineer – Computer Vision & Video

Objectways; Chennai District; In loco; INR 1.500.000 - 2.500.000; Tempo pieno

Descrizione del lavoro

Objectways is seeking a senior Applied AI/ML Engineer to lead production-grade AI solutions for video understanding, detection/segmentation, tracking, and multimodal reasoning. You will take problems from data to deployment, with a focus on practical research and engineering in vision/video ML.

The role emphasizes end-to-end model ownership, dataset design, and GPU-accelerated inference, collaborating with backend/MLOps and annotation teams to deliver robust production services.

Competenze

• Proven hands-on experience in applied AI/ML for CV/video tasks.
• End-to-end model ownership from data gathering to deployment.
• Experience with datasets, evaluation protocols and leakage prevention.

Mansioni

• Build, fine-tune and benchmark models for object detection, segmentation, tracking, and video understanding.
• Design datasets, annotation schemas, gold sets and leakage-safe evaluation protocols; perform failure analysis.
• Evaluate modern vision/video/multimodal models and decide on pipelines; document decisions.

Conoscenze

Python

Computer Vision

Video ML

Multimodal Systems

Experiment Design

End-to-End Ownership

Leadership

Formazione

B.Tech / M.Tech / MS / PhD in CS/AI/Data Science/Engineering/Robotics

Strumenti

OpenCV

FFmpeg

Docker

FastAPI

CUDA

Descrizione del lavoro

We are looking for a senior Applied AI/ML Engineer with strong hands‑on experience in Computer Vision and video systems to help lead the development of production‑grade AI solutions for video understanding, detection/segmentation, tracking, multimodal reasoning and model‑assisted annotation. This is a practical research‑and‑engineering role: the person should be comfortable taking a problem from data and experimentation through model training, evaluation, optimization and deployment. Prior Physical AI or robotics experience is not mandatory, but the candidate must have enough depth in vision/video ML to grow into those areas.

Roles and Responsibilities

• Build, fine‑tune and benchmark models for areas such as object detection, instance/semantic segmentation, tracking, video understanding, action recognition or other temporal vision tasks.
• Design reliable datasets, annotation schemas, gold sets and leakage‑safe evaluation protocols, and perform systematic failure analysis across real‑world video conditions.
• Evaluate modern vision, video and multimodal models and decide when to use a general‑purpose model, a specialist model, or a hybrid pipeline.
• Optimize GPU inference and integrate versioned models into production services, working with backend/MLOps and annotation‑platform teams.
• Own or co‑own the technical design of applied AI/ML solutions with a strong Computer Vision and video component, from problem definition and data strategy through deployment.

Profile:

We are open to experienced Applied AI/ML candidates whose strongest work is in Computer Vision, video ML or multimodal systems. The exact previous job title is not important; demonstrated end‑to‑end ownership of real model‑development work is.

• Relevant sourcing profiles / titles include: Senior Applied AI/ML Engineer; Senior AI/ML Engineer – Vision & Multimodal Systems; Senior Computer Vision Engineer; Machine Learning Engineer – Computer Vision; Video AI Engineer; Applied Scientist – Computer Vision; Multimodal AI Engineer; and Applied ML Research Engineer.
• Typically 4–8+ years of relevant hands‑on experience in applied AI/ML, with a meaningful portion of that work involving Computer Vision or video. Strong candidates with fewer years but clear end‑to‑end model ownership may also be considered.
• B.Tech / M.Tech / MS / PhD in Computer Science, AI, Data Science, Electrical/ECE, Mathematics & Computing, Engineering Physics, Robotics, Mechatronics or a related technical discipline. Equivalent demonstrated industry or research depth is also acceptable.

Required Qualifications:

• Strong Python and hands‑on PyTorch experience, including model training, fine‑tuning, debugging and experiment design.
• Solid practical experience in Computer Vision with real model‑development work across at least two of the following: object detection, segmentation, tracking, pose estimation, video understanding, action recognition/localization or related temporal vision problems.
• Hands‑on experience processing real image/video data using OpenCV, FFmpeg or equivalent tooling, including frame/clip preparation, augmentation and data‑quality handling.
• Ability to design datasets and evaluation protocols correctly: train/validation/test construction, leakage prevention, class imbalance, hard negatives, gold‑set creation and reproducible benchmarking.
• Strong model‑evaluation and failure‑analysis skills; able to reason about errors caused by motion blur, occlusion, domain shift, lighting, viewpoint changes, tracking drift or poor segmentation boundaries.
• Experience training or fine‑tuning deep‑learning models on GPUs, with practical understanding of VRAM, batching, mixed precision, throughput and inference trade‑offs.
• Experience taking models beyond notebooks: Docker‑based deployment and API/model serving using FastAPI or equivalent, with practical cloud/GPU deployment exposure.
• Ability to compare multiple technical approaches rather than simply integrate a single popular model, and to explain architecture decisions using measurable quality, latency and reliability trade‑offs.
• Ability to provide technical guidance to junior engineers while remaining hands‑on with code, experiments and debugging.

Preferred Qualifications:

• Multimodal/VLM or open‑vocabulary vision experience, including vision‑language grounding, video‑language understanding or multimodal retrieval.
• Egocentric/first‑person video, Physical/Embodied AI, hand‑object interaction, manipulation data or annotation‑automation experience.
• Experience with modern model families or equivalent approaches such as Grounding DINO, SAM‑style models, Florence, Qwen‑VL, InternVideo, ActionFormer, Mask2Former or contemporary detection/segmentation/video architectures.
• ONNX, TensorRT, Triton Inference Server, CUDA/performance profiling, distributed training or multi‑GPU inference.
• Experience integrating model outputs into human‑in‑the‑loop annotation platforms such as TensorAct, Encord, CVAT, Label Studio or equivalent.