About BigLook
BigLook is a commercialisation project at Trinity College Dublin, embedded within the ADAPT Centre. The project is focused on the research and development of AI-driven art intelligence technology. We are advancing the state of the art in multimodal artificial intelligence for the cultural sector, focused on developing a "System of Work" for the global art market powered by a domain-specific Large Multimodal Model (LMM) and a suite of AI Agents. Our work integrates cutting-edge research in computer vision, natural language processing, vision-language models, generative AI, and reinforcement learning with human feedback, translating these into production-grade systems capable of nuanced visual analysis, metadata extraction, and generative interpretation of artworks. Collaborating closely with art historians, galleries, and collectors, we combine rigorous AI research with domain expertise to produce robust, high-fidelity models that deliver novel insights into visual creativity.
About the role
The Research AI Developer will architect and implement a domain-specialized LMM for the art domain, integrating high-resolution imagery, textual metadata, and instructional data via advanced transformer-based architectures to produce semantically aligned joint embeddings. The system will support expert-level visual analysis, fine-grained attribute extraction, taxonomy classification, personalised recommendations, and context-aware generative tasks.
You will apply state-of-the-art generative models (diffusion, GANs, transformer decoders) with domain-aware conditioning, alongside transfer learning, meta-learning, and few/zero-shot strategies to handle heterogeneous styles, rare classes, and evolving art taxonomies. The role involves fine-tuning vision-language transformers using parameter-efficient techniques (LoRA, adapters), implementing hierarchical and cross-modal attention, and optimising multimodal alignment losses for semantic coherence. RLHF pipelines will be developed to align outputs with expert evaluative heuristics.
You will also be responsible for architecting and implementing sophisticated autonomous agents that can reason, plan, and execute complex tasks across the art market ecosystem. This involves designing agents capable of dynamic goal formulation, strategic task decomposition, and adaptive learning in response to real-time market data. A key focus will be on developing novel algorithms for agent communication and coordination, exploring techniques beyond standard protocols to enable emergent collaborative behaviors.
This position combines foundational multimodal AI research with production-grade deployment, delivering a robust, high-fidelity LMM optimised for subjective, high-variance visual art interpretation and creation.
Key Responsibilities
* Architect and Advance Domain-Specific LMMs
– Design, pre-train, and fine-tune state-of-the-art multimodal architectures (e.g., BLIP-2, Llama 4) for deep art analysis, integrating high-resolution imagery, structured/unstructured metadata, and provenance records into unified semantic embeddings.
* Generative AI Pipelines for Art
– Develop advanced generative systems leveraging diffusion models, StyleGAN variants, and autoregressive transformer decoders with domain-specific conditioning for style transfer, restoration, synthesis, and augmentation of artworks.
* Adaptive Learning for Rare and Evolving Styles
– Implement domain adaptation, meta-learning, and few/zero-shot generalisation strategies to support emerging art demands, categories, and ontologies in the art market.
* Efficient and Scalable Fine-Tuning
– Employ parameter-efficient training to adapt foundation LMM/LLM models to the art domain while optimising GPU/TPU resource utilisation.
* Human-Aligned Multimodal Reasoning
– Design and run RLHF pipelines and preference-modelling loops that incorporate feedback from art historians, curators, and valuation experts, ensuring subjective alignment in interpretive tasks.
* Attention Optimisation and Multiscale Contextual Modelling
– Engineer hierarchical, cross-modal, and sparse attention mechanisms (e.g., Performer, LongNet) to capture fine-grained visual detail and contextual relationships across large, heterogeneous datasets.
* Research-to-Product Translation
– Rapidly prototype, benchmark, and productionise research innovations, bridging SOTA academic models with deployable APIs and scalable MLOps pipelines optimised for low-latency, high-fidelity outputs.
* Evaluation Beyond Accuracy
– Establish hybrid evaluation frameworks combining quantitative metrics (BLEU, CIDEr, CLIPScore, FID) with qualitative expert panel reviews, cultural expert checks, and subjective aesthetic scoring.
* Multi-agent framework
– Design and implement a scalable, asynchronous multi-agent framework, creating robust systems for agent communication, task delegation, and coordinated execution.
* Cross-Disciplinary Collaboration
– Work with art experts to define ontology schemas, annotation protocols, and evaluation rubrics that guide both supervised learning and self-supervised representation learning pipelines.
Essential Qualifications & Experience
* At least 4 years of relevant experience in AI, machine learning, or data science.
* Strong background in deep learning and computer vision. Proficiency in Python and ML frameworks (e.g., PyTorch, TensorFlow).
* Demonstrated experience with transformer-based architectures and attention mechanisms.
* Experience with vision-language or multimodal models (e.g., CLIP, Vision Transformers, GPT-4V) and generative models (GANs, diffusion).
* Familiarity with advanced fine-tuning methods (LoRA), RLHF techniques, and few-shot/zero-shot learning paradigms.
* Familiarity with multi-agent system frameworks such as LangGraph, AutoGen, or CrewAI.
* Solid understanding of agentic design patterns, planning algorithms, and inter-agent communication protocols.
* Creative problem-solving skills & a track record of innovation (publications or projects).
* Teamwork abilities for collaborating in a multidisciplinary startup environment.
* Master's or PhD in Computer Science, Artificial Intelligence, or comparable commercial experience.
All applicants should have a valid work permit or legal right to work in Ireland.