Jobs at ifm

Conduct and prototype novel reinforcement learning research for foundation models, including massive-scale self-play, end-to-end engineering from data curation to production, contribute to training/inference frameworks, publish and present results, engage with open-source and external collaborators.

Manage end-to-end HPC infrastructure projects for large-scale GPU clusters: coordinate vendors and internal stakeholders, track milestones and risks, maintain documentation and governance, support compliance and export-control tracking, and produce executive reports and dashboards to ensure research timelines are met.

Drive research and engineering to develop PAN world models: scalable training on large clusters, video data pipelines and annotation, evaluation benchmarks, visual tokenization, and inference optimization for real-time interaction.

Short-term intensive research internship assisting in dataset collection and preprocessing, model architecture design and tuning, implementation of testing frameworks, analysis of model accuracy and bias, algorithm refinement, and documentation to support foundation-model research.

Lead research to improve LLM efficiency and architectures, design and implement NLP algorithms (using PyTorch), integrate LLMs with vision models, optimize models and data pipelines, collaborate on code/design reviews, publish technical reports and papers, represent the institute at conferences, and contribute to ML infrastructure and model evaluation.

Coordinate cross-functional work across research, engineering, product, marketing, and community for the Abu Dhabi lab. Translate priorities into project plans, track roadmaps and readiness milestones, run end-to-end model launches, liaise with MarComms and partners, maintain lightweight processes, report progress and risks to leadership, and drive resolution of blockers while improving launch and delivery processes.

Design and optimize communication stacks and runtime for large-scale distributed training (including MoE and hybrid parallelism). Improve collective performance, reduce latency, ensure fault-tolerant execution, debug NCCL/UCX/RDMA issues, and co-design topology-aware orchestration for 1,000+ GPU clusters.

Rapidly collect, curate, and preprocess high-quality NLP datasets. Build and maintain web crawlers, APIs, automated workflows, and scalable data pipelines. Refine and evaluate LLM outputs, ensure data quality and performance, document methodologies, and collaborate with researchers and engineers to deliver data for model training and benchmarking.

Develop and maintain backend services (Python/FastAPI), data pipelines, and RESTful APIs; build frontend web apps (Vue/React/Next) and Flutter mobile apps; integrate model inference and data visualization; manage PostgreSQL databases; implement CI/CD, Docker, and cloud deployment; ensure performance, security, and cross-team collaboration.

Own end-to-end agentic data pipelines and evaluation suites: source and curate trajectories and tool-use data, design RL environments and rewards, build reliable large-scale eval harnesses, run experiments, diagnose regressions, and produce research outputs and technical reports.

Support the Paris IFM lab with day-to-day finance and administrative operations: payroll processing, budget monitoring, procurement, vendor relations, IT coordination, compliance with MBZUAI and French regulations, recordkeeping, travel coordination, and event support to enable researchers to focus on scientific work.

Build and grow a developer and user community around IFM foundation models; run events and trainings; translate research into docs, demos, and onboarding; support external adopters and gather feedback to inform product and research roadmaps.

Design, extend, and maintain distributed training infrastructure: modify frameworks (DeepSpeed, FSDP, etc.), implement distributed optimizers, build multi-node launch/config systems, implement experiment tracking and monitoring, optimize reliability and performance, and write production-quality ML infra code in PyTorch or JAX.

Lead product vision and strategy for foundation model families (LLMs, reasoning, world models). Define user personas and roadmaps, translate research into production requirements, prioritize features, own release criteria, drive cross-functional execution, and build/scale the Product Management organization while ensuring performance and safety for public releases.

Build and maintain scalable multimodal (primarily video) data pipelines, support large-scale distributed training and inference optimization for world and video generative models, integrate research prototypes into production, ensure system reliability, and contribute to evaluation, documentation, and research outputs.

Intern will build and validate a simulator and profiling framework to model inference on NVIDIA GPUs, develop analytical GPU kernel performance models, profile kernels with Nsight tools, analyze PTX/SASS, identify bottlenecks, and recommend kernel and system-level optimizations for transformer inference on Hopper and Blackwell architectures.

Lead research on efficient large language models and multimodal integration. Design, implement, and optimize NLP algorithms and training; collaborate across teams, write technical reports and papers, and represent the institute at conferences.

Research and engineer agentic language models: develop algorithms, datasets, evaluations, and scalable RL and distributed training/serving infrastructure; publish papers and contribute open-source software.

Build and productionize distributed pre-training stacks (PyTorch/JAX) at multi-node GPU scale, implement new optimizers and high-performance CUDA/Triton kernels, optimize mixed-precision training and memory/communication, develop experiment infrastructure and monitoring, and mentor junior engineers to accelerate foundational model research.

Build and operate Eval360 evaluation service, perform deep error analysis on model outputs, design error taxonomies and review interfaces, implement front-end and back-end evaluation pipelines, dashboards, APIs, and data workflows, and collaborate with researchers and engineering teams to improve model quality and evaluation infrastructure.