Applied AI Researcher (Brazil)

About the role

Articul8, a GenAI platform serving regulated industries such as energy, semiconductors, finance, aerospace and supply chain, is looking for an Applied AI Researcher to advance its domain-specific GenAI platform. The role covers model training, reinforcement learning, multimodal understanding and knowledge representation, and involves designing and running experiments, building training and evaluation pipelines, and shipping research into production.

Key responsibilities

• Architect and orchestrate massively parallel AI research workflows, using fleets of agentic AI systems to explore hypotheses, hyperparameters and architectures at scale.
• Design, train and iterate on models across the GenAI stack (LLMs, VLMs, embedding models, rerankers, reward models) with agentic pipelines that manage data preprocessing, training runs, evaluation sweeps and result synthesis.
• Conduct first-principles research on model architectures, training dynamics, reinforcement learning and knowledge representation, using AI agents to accelerate literature reviews, ablation studies and mathematical analysis.
• Span disciplines and modalities (NLP, computer vision, multimodal understanding, agentic reasoning, domain science) by delegating exploration, prototyping and benchmarking to parallel agents.
• Build agentic research infrastructure: tooling, libraries and platforms that let researchers orchestrate autonomous experiment pipelines, data workflows and evaluation harnesses.
• Ship research into production with engineering, product and domain experts, using agentic CI/CD and automated integration testing.
• Document findings, publish at top-tier venues and build internal knowledge systems that agentic tools can index.
• Identify bottlenecks in your own and the team's workflow and design or adopt scalable solutions.

Required profile

• PhD in Computer Science, Machine Learning or a related field, or MSc with 4+ years of post-graduation research experience.
• Has trained or fine-tuned at least one neural model end-to-end, from data preparation to evaluation.
• Strong working knowledge of probability, optimization and linear algebra applied to NLP, computer vision, reinforcement learning or information retrieval.
• Experience building training or evaluation pipelines with real data: preprocessing, distributed computation, experiment tracking and reproducibility.
• Production-quality Python, fluent with Git and collaborative development workflows.

Preferred qualifications

• Distributed training frameworks (PyTorch DDP, DeepSpeed, FSDP).
• Papers published at NeurIPS, ICML, ICLR, ACL, EMNLP, CVPR or equivalent.
• Hands-on experience with RLHF, DPO and reward modeling.
• Cloud infrastructure experience (AWS, GCP or Azure) for ML workloads.