Multimodal Research Scientist

We're AtkinsRéalis, a world class Engineering Services and Nuclear organization. We connect people, data and technology to transform the world's infrastructure and energy systems. Together, with our industry partners and clients, and our global team of consultants, designers, engineers and project managers, we can change the world. Created by the integration of long-standing organizations dating back to 1911, we are a world-leading professional services company dedicated to engineering a better future for our planet and its people. We deploy global capabilities locally to our clients and deliver unique end-to-end services across the whole life cycle of an asset including consulting, advisory & environmental services, intelligent networks & cybersecurity, design & engineering, procurement, project & construction management, operations & maintenance, decommissioning and capital. The breadth and depth of our capabilities are delivered to clients in key strategic sectors.

News and information are available at or follow us on LinkedIn.

Our teams take great pride in delivering some of the world's most prestigious projects. This success is driven by our talented people, whose diverse perspectives, expertise, and knowledge set us apart. Join us and you'll be part of our genuinely collaborative environment, where everyone is supported to make the most of their talents and expertise.

When it comes to work-life balance, AtkinsRéalis is a great place to be. So, let's discuss how our flexible and remote working policies can support your priorities. We're passionate about are work while valuing each other equally. So, ask us about some of our recent pledges for Women's Equality and being a 'Disability Confident' and 'Inclusive Employer'.

You'll be the AI/ML Lead on a small, cross-functional team. Your mission:

• Build a custom image generation + editing model tailored to Urban, architecture and interior design.
• Develop a pipeline for image understanding code generation (For industry standard software's)
• Train or fine-tune a Small Language Model (SLM) on domain-specific workflows, APIs, and design knowledge.
• You'll work closely with our Systems Engineer (who handles integrations and data pipelines) and our Product/Domain Lead (who supplies domain expertise, data, and pilot users).

Key Responsibilities

• Model Design & Training:

• Build and train a diffusion or transformer-based text-to-image model with text-driven editing (ControlNet/LoRA/inpainting) for architecture and interiors.

• Combine a vision encoder with a code/SLM decoder to output parametric tool plans and executable code for industry software.

• SLM Fine-Tuning & Program Synthesis

• Select and fine-tune an open-source LLM/SLM (Llama, Qwen, DeepSeek, etc.) to encode architecture and interior design workflows and APIs.

• Design structured outputs, grammar constraints, and execution-guided training for high tool-call accuracy.

• Data Strategy

• Develop large, high-quality datasets pairing images, prompts, and gold-standard tool calls/code.

• Lead annotation/auto-labeling strategies and data augmentation tailored to architecture/interior contexts.

• Collaboration & Integration

• Partner with the Systems Engineer to ensure your models integrate smoothly with Industry standard softwares

• Incorporate domain feedback from the Product/Domain Lead and pilot users to guide model improvements.

• Research & Innovation

• Stay ahead of the state of the art in generative models, program synthesis, and multimodal LLMs.

• Prototype novel architectures for latent scene representation, spatial reasoning, and text-based editing.

Qualifications:

• Master's or PhD in Computer Science, Machine Learning, Computer Vision, or related field.
• Experience: 4–5 years of hands-on research and development in generative models, computer vision, or program synthesis.

• Technical Skills:

• Expert in PyTorch or TensorFlow, plus supporting libraries for vision and NLP.

• Proven experience training diffusion models or large transformer architectures at scale.
• Comfort with fine-tuning LLMs (LoRA/QLoRA) and structured decoding (JSON/grammar constraints).
• Familiarity with 3D or CAD/BIM concepts — at least enough to design good intermediate representations.

What We Can Offer You:

• Varied, interesting and meaningful work.
• A hybrid working environment with flexibility and great opportunities.
• Opportunities for training and, as the team grows, career progression or sideways moves.
• An opportunity to work within a large global multi-disciplinary consultancy on a mission to change the ways we approach business as usual.

Why work for AtkinsRéalis?

We at AtkinsRéalis are committed to developing its people both personally and professionally. Our colleagues have the advantage of access to a high ranging training portfolio and development activities designed to help make the best of individual's abilities and talents. We also actively support staff in achieving corporate membership of relevant institutions.

Meeting Your Needs:

To help you get the most out of life in and outside of work, we offer employees 'Total Reward'.

Making sure you're supported is important to us. So, if you identify as having a disability, tell us ahead of your interview, and we'll discuss any adjustments you might need.

Additional Information:

We are an equal opportunity, drug-free employer committed to promoting a diverse and inclusive community - a place where we can all be ourselves, thrive and develop. To help embed inclusion for all, from day one, we offer a range of family friendly, inclusive employment policies, flexible working arrangements and employee networks to support staff from different backgrounds. As an Equal Opportunities Employer, we value applications from all backgrounds, cultures and ability.

We care about your privacy and are committed to protecting your privacy. Please consult our Privacy Notice on our Careers site to know more about how we collect, use and transfer your Personal Data.

Link: Equality, diversity & inclusion | Atkins India )