Role Overview
A client of bytespark.ai is seeking a Principal Applied AI Architect to lead the strategy, architecture, and delivery of advanced AI solutions for high-volume voice, text, and multimodal data processing. The successful candidate will own end-to-end technical decisions for large language models, speech recognition, diffusion models, and production machine learning platforms. This role will design and fine-tune transformer-based systems for classification, entity extraction, sentiment analysis, conversational AI, and other domain-specific applications. The architect will establish scalable approaches for distributed training, model and hardware sizing, parameter-efficient fine-tuning, inference optimization, and real-time deployment. They will guide the development of robust pipelines that process billions of data points while meeting demanding reliability, security, governance, and performance expectations. The position will partner with engineering, product, research, and operational stakeholders to translate complex requirements into maintainable AI architectures and delivery roadmaps. Responsibilities include defining evaluation frameworks, monitoring model quality, directing A/B testing, and promoting reproducible, explainable, and ethical AI practices. The architect will document standards, mentor machine learning engineers, review critical designs, and advance engineering practices across multiple teams. This is an opportunity for a hands-on technical leader to shape consequential AI systems while continuously evaluating emerging research and technologies.
Requirements
- Advanced degree in Computer Science, Machine Learning, Data Science, Mathematics, or a related discipline, with at least 4 years of machine learning engineering experience focused on deep learning and neural networks.
- Principal-level experience defining end-to-end applied AI architecture, technical strategy, governance standards, design patterns, and delivery roadmaps for high-impact production programs.
- Expert Python proficiency and advanced hands-on experience with PyTorch, TensorFlow or JAX, HuggingFace Transformers, and at least one performance-oriented language such as C, C++, JavaScript, or Julia.
- Demonstrated delivery of transformer and LLM solutions using architectures such as GPT, BERT, T5, or LLaMA, including fine-tuning for NLP and conversational AI use cases.
- Production experience with ASR and speech-processing pipelines using technologies such as Whisper, Wav2Vec2, torchaudio, librosa, or SpeechBrain.
- Experience sizing, optimizing, and deploying large models using distributed training, model parallelism, quantization, pruning, distillation, ONNX, TensorRT, or comparable methods.
- Ability to build distributed multimodal pipelines and operate models with Docker, Kubernetes, MLflow, Weights & Biases, cloud ML platforms, Git, DVC, Spark, Dask, or Ray.
- Strong technical leadership, English communication, documentation, stakeholder collaboration, design review, and mentoring capabilities.
Desirable
- Hands-on experience with diffusion models, VAEs, GANs, synthetic data generation, and multimodal content generation.
- Knowledge of RLHF, constitutional AI, federated learning, privacy-preserving ML, explainable AI, bias detection, or adversarial machine learning.
- Experience building real-time inference and streaming architectures for telecommunications, public-sector, cybersecurity, or similarly regulated environments.
- Published AI research, patents, conference participation, or meaningful contributions to open-source machine learning projects.
- Experience with edge deployment, mobile model optimization, graph neural networks, knowledge graphs, AutoML, or neural architecture search.
This vacancy remains open and applications are being reviewed. Originally posted .