R&D AI Engineer

About the Company Atna.ai is seeking a self-directed R&D AI Engineer with 2-3 years of dedicated, hands-on experience manipulating foundation models. In this role, you will take ownership of our synthetic media detection initiatives, dissecting state-of-the-art neural network architectures and customizing their underlying components to solve complex classification and segmentation challenges. As an independent researcher and solo contributor, you will bridge the gap between cutting-edge theoretical research and scalable, real-world deployment. About the Role In this role, you will take ownership of our synthetic media detection initiatives, dissecting state-of-the-art neural network architectures and customizing their underlying components to solve complex classification and segmentation challenges. Responsibilities - Architect, modify, and fine-tune foundation models for deepfake image detection, image-to-image edit segmentation, and complex video/audio manipulation analysis. - Deconstruct and rebuild core architectural components—specifically Transformer encoders/decoders, multi-head attention blocks, and latent space embeddings—rather than relying on out-of-the-box API implementations. - Implement and iterate upon CLIP-based multimodal models, Vision Transformers (ViT), and advanced U-Net architectures for cross-modal forensic analysis and high-precision spatial segmentation. - Engineer specialized loss functions (e.g., combined BCE + Dice, contrastive loss) and optimize gradient flows to achieve subpixel boundary accuracy in synthetic media detection. - Refactor experimental PyTorch research code into highly optimized, production-ready Python pipelines, ensuring strict dependency management and CPU/GPU inference optimization. - Containerize AI microservices using Docker and manage seamless deployments across multi-node Swarm clusters utilizing GitHub Actions CI/CD pipelines. Qualifications - Experience: 3-5 years of applied experience specifically training, fine-tuning, and modifying foundation models. - Core ML/DL Stack: Deep proficiency in Python, PyTorch, NumPy, and CUDA tensor optimization. - Architectural Mastery: Intimate knowledge of Vision Transformers (ViT), CLIP embeddings, Auto-Encoders, U-Net, and self-supervised learning (SSL) paradigms. - Deployment & Infrastructure: Strong capabilities in Python environment management (e.g., Poetry), Docker, Docker Swarm, GitHub Actions, and RESTful API development (Flask/FastAPI) for high-throughput, cluster-based environments. - Advanced Techniques: Expertise in latent space manipulation, transfer learning, principal component analysis (PCA), and integrating advanced algorithmic logic (e.g., graph-based shortest-path algorithms) with neural network probability maps. - Research Acumen: Demonstrated ability to autonomously read, comprehend, and translate complex AI research papers into performant, scalable code without external guidance. Required Skills - Deep proficiency in Python, PyTorch, NumPy, and CUDA tensor optimization. - Intimate knowledge of Vision Transformers (ViT), CLIP embeddings, Auto-Encoders, U-Net, and self-supervised learning (SSL) paradigms. - Strong capabilities in Python environment management (e.g., Poetry), Docker, Docker Swarm, GitHub Actions, and RESTful API development (Flask/FastAPI). Preferred Skills - Expertise in latent space manipulation, transfer learning, principal component analysis (PCA). - Integrating advanced algorithmic logic (e.g., graph-based shortest-path algorithms) with neural network probability maps.