Senior Applied AI Researcher (Dublin, CA)

• Own and orchestrate end-to-end research programs using massively parallel agentic AI — from problem formulation through production deployment, designing agent-driven experiment campaigns that simultaneously explore model architectures, training regimes, data strategies, and evaluation criteria at a pace and breadth that redefines what a single researcher can accomplish
• Go deep: drive breakthrough domain-specific model quality — lead multi-stage training pipelines, domain adaptation, RL-based optimization (RLHF, DPO, reward modeling), and training dynamics analysis, using agentic systems to run exhaustive ablations, hyperparameter sweeps, and failure-mode investigations in parallel
• Go broad: span modalities, methods, and domains simultaneously — design and train multimodal systems (text, images, tables, charts, technical documents), knowledge graph pipelines, hybrid retrieval architectures, and structured reasoning systems, delegating exploration and prototyping across these fronts to parallel agent workflows so you can synthesize cross-cutting insights in real time
• Architect agentic data and training infrastructure — build agent-orchestrated pipelines for domain-specific data curation, quality filtering, preprocessing, and large-scale training that the entire research team can leverage to go faster
• Mentor AI Researchers in the agentic paradigm — coach team members on how to amplify their own depth and breadth by designing effective agent workflows, raising the ceiling on what every researcher can achieve
• Compress the research-to-production cycle — take prototypes to production-ready systems rapidly by leveraging agentic CI/CD, automated integration testing, and continuous evaluation harnesses, collaborating closely with engineering, product, and domain experts
• Build force-multiplying knowledge systems — document findings, publish at top-tier venues, and contribute to internal knowledge infrastructure that agentic tools can index and reason over, turning every breakthrough into compounding team-wide leverage
• Model the augmented researcher – continuously identify bottlenecks in your own and the team's workflows, then design or adopt efficient, scalable solutions that eliminate them — treating the maximization of human potential as a first-class research output

• Education: PhD or MSc in Computer Science, Machine Learning, or a related field.
• Experience: 5+ years as an AI/ML researcher with shipped research artifacts (models, systems, or tools in production), including 2+ years building LLM-based systems.
• Model training depth: You have run multi-stage training pipelines (pretraining, fine-tuning, post-training) and can diagnose training failures from loss curves, gradient norms, and evaluation metrics — not just restart the job.
• Technical specialization: Deep expertise in at least one of: domain-specific model adaptation, multimodal learning, reinforcement learning from human feedback, knowledge-grounded generation, or retrieval-augmented systems. You've published or shipped production work in your area.
• Distributed systems: Hands-on experience with distributed training at scale (DeepSpeed, FSDP, Megatron-LM, or equivalent). You understand data parallelism vs. model parallelism and know when each matters.
• Software engineering: Production-grade Python, clean abstractions, tested code. You build tools others depend on.

• Experience adapting models to specialized domains where standard benchmarks don't apply — you've had to define what "correct" means and build evaluation around it.
• Track record of taking a research prototype to a production system serving real users.
• Experience with knowledge graph construction, hybrid retrieval architectures, or structured reasoning systems in practice — not just in papers.
• Strong publication record with evidence of depth, not just breadth.
• Cloud-native ML infrastructure experience (Kubernetes, distributed job scheduling, GPU cluster management).

Professional Attributes (Code42)

• Practice Humility: You stay open to being wrong — even on problems you've studied for years. You actively seek perspectives that challenge your assumptions and create space for junior researchers to teach you something new.
• Bias for Outcomes: You own the outcome end-to-end, from research question to production impact. You know the difference between interesting work and important work, and you choose the latter. Dates and scope are commitments, not suggestions.
• Care Deeply: You treat your team's problems as your own. You give honest, specific feedback because you care about the person's growth, not just the project. When something is broken, you fix it or flag it — you never walk past it.
• Dare to Do the Impossible & Embrace Scarcity: You pursue research directions others have written off. You use resource constraints as forcing functions for creative solutions, not excuses. Your ambition is calibrated to what the problem demands, not what feels safe.
• Build a Better World: You hold yourself to the standard that your work should make the enterprise smarter, not just the model better. You mentor others because raising the bar for the whole team is how you multiply impact.

Articul8 was born from a simple belief: GenAI should work for the enterprise, not the other way around. Our platform — combining domain-specific models, autonomous agentic reasoning (ModelMesh™), reliable model evaluation (LLM-IQ™), and multimodal understanding — serves regulated industries such energy, semiconductor, finance, aerospace, supply chain, and more. Trusted by Fortune 500 enterprises, we bring together research, engineering, product, and domain expertise to deliver AI that meets the accuracy, explainability, and auditability standards that high-stakes environments demand.