DSML.KZ Community Hub

Dynamic and innovative AI Tech environment. Responsibilities: • Design, train, fine-tune, evaluate, and implement generative models for applications with images, video, text, and multimodal data • Improve visual quality of image and video generation: prompt adherence, spatial alignment, character and object consistency, temporal consistency • Research and implement methods for controlling LLM: instruction tuning, structured generation, constrained decoding, tool usage, safety mechanisms, preference optimization • Create evaluation systems for LLM: instruction execution, reliability, factual accuracy, tool usage accuracy, robustness against adversarial prompts • Optimize training and inference pipelines for latency, throughput, memory usage, and infrastructure costs • Build scalable data processing, training, fine-tuning, and real-time inference pipelines • Integrate generative models into production systems in collaboration with engineers • MLOps: automate testing, deployment, monitoring, rollback, and model retraining • Analyze recent research in generative AI and assess its practical value Requirements: • At least 3 years of experience in machine learning and software development • Extensive hands-on experience in developing/deploying generative AI models • Deep understanding of diffusion models and image/video generation architectures • Experience with LLM: prompting, fine-tuning, evaluation, structured outputs, tool invocation • Hands-on experience with PyTorch or TensorFlow • Experience transitioning models from prototype to production, operating inference services • Experience deploying models on GPU servers and cloud platforms • Strong Python skills and software development: testing, monitoring, debugging, optimization Optional: • Experience with RLHF, DPO, reinforcement learning, instruction tuning • Experience with constrained decoding, function calling, agent systems, red-teaming • Experience with multimodal models (text, image, video, audio) • Improving temporal consistency, spatial alignment, identity preservation in video • Optimization of GPU computations: quantization, batching, caching, compilation • Experience with MLOps platforms, model registries, experiment tracking • Familiarity with CI/CD and Infrastructure as Code • Experience with high-throughput low-latency systems