About the Company and Project

We are TechnoForce. We are launching a new product project HODAR. The product is a video analytics system for monitoring personnel activity at customer sites. The solution is designed for enterprises with a distributed workforce and automates time tracking, detection of regulatory violations, and personnel workload assessment based on processing video streams from surveillance cameras.

Currently, a development team for the project is being formed. The work will cover all stages of the lifecycle: from data analysis to industrial implementation and subsequent support.

Areas of Work within the Project:

• Initial analysis of video data (suitability assessment, artifacts, class statistics, selection of augmentation and annotation strategies).
• Development and training of neural network models for detection, tracking, and action classification.
• Optimization of model computational efficiency for real-time operation on limited hardware resources.
• Building a video stream processing pipeline and integration with the product's backend infrastructure.
• Production model support: monitoring quality metrics, retraining, handling data drift.

Job Responsibilities

• Conducting research analysis of video data: assessing resolution, bitrate, frame rate, lighting conditions, noise, object distribution.
• Implementing detection and tracking algorithms using PyTorch/TensorFlow frameworks (YOLO, Detectron2, DeepSORT, ByteTrack).
• Developing software modules for capturing, decoding, and preprocessing video streams (OpenCV, FFmpeg, multithreading).
• Applying inference optimization methods: quantization, pruning, conversion to ONNX/TensorRT, reducing latency while maintaining accuracy.
• Writing high-performance Python code using vectorized computations (NumPy) and asynchronous approaches.
• Integrating ML components with external systems via REST API or message queues.
• Documenting adopted architectural decisions, experiment results, and deployment instructions.
• Participating in discussions on data requirements and quality metrics with customers and analysts.

Candidate Requirements

• Commercial experience in computer vision and video analytics — from 2 years.
• Fundamental mathematical background: linear algebra, probability theory, mathematical statistics. Understanding of gradient optimization methods, loss function structures (Cross-Entropy, Focal Loss, IoU Loss) and their modifications. Ability to reasonably select and adapt algorithms to specific video task characteristics.
• Practical proficiency in Python and specialized libraries:
  • Computer vision and deep learning libraries: OpenCV, PyTorch, TensorFlow, Detectron2, YOLO (v5–v10), DeepSORT, MediaPipe.
  • Data analysis and visualization tools: Pandas, NumPy, SciPy, Matplotlib, Seaborn.
  • Video processing tools: FFmpeg (capture, conversion, frame sampling).
• Experience in model performance optimization (TensorRT, OpenVINO, ONNX Runtime).
• Containerization and deployment skills (Docker, FastAPI/Flask, working with PostgreSQL/ClickHouse databases, Redis/Kafka message brokers).
• Knowledge of video transmission protocols (RTSP, H.264, H.265) and video data formats.
• Higher technical education (applied mathematics, computer science, cybernetics, radiophysics, or related fields) — mandatory.
• Independence, ability to formulate tasks and propose reasoned solutions, responsibility for model performance in operation.
• Version control system skills (Git) and team development practices (code review, CI/CD — welcome).

Working Conditions

• Participation in product development from scratch, ability to influence architecture and technology stack.
• Remote work format (fully or partially — negotiable).
• Absence of excessive bureaucracy, focus on engineering and research tasks.
• Salary based on interview results, in line with market rates.
• Employment under the Labor Code of the Russian Federation or under a civil law contract.