Postdoctoral Researcher – Generative AI Agents (SCOPE Lab)

Sobre a vaga

Offer Description The Champalimaud Foundation (Fundação D. Anna de Sommer Champalimaud e Dr. Carlos Montez Champalimaud), a private, non-profit research institution in Lisbon, Portugal, is looking for a Postdoctoral Researcher to join the Surgery, Care, Outcomes, Personalization, and Empowerment (SCOPE) Lab, within the Breast Cancer Research Programme. The SCOPE Lab is led by Professor Maria João Cardoso (Head of the Breast Unit, Champalimaud Foundation). The postdoctoral researcher will be assisted in their scientific and technical work by Dr. João Santinha (PI and Co-Lead of the Digital Surgery Lab and member of the Breast Imaging Group) and Dr. Luís Elvas (Health Data Engineer of the Breast Cancer Research Programme). The selected candidate will lead Champalimaud Foundation´s technical contributions to ResPECT (Representing People´s Experience of Cancer and its Treatment) - one Pan-European generative AI project in healthcare, funded with €16 million by the Horizon Europe programme (HORIZON-HLTH-2025-01-CARE-01). ResPECT brings together leading partners from over 10 countries to develop the first end-user–driven, trustworthy generative AI Virtual Assistant for patients, clinicians, and health systems in oncology and mental health care. CF leads Work Package 3 (Generative AI agent and system architecture) of the project, with key responsibilities also across WP4 (Medical and Policy Assistants), WP5 (LLM output validation), and WP7 (legal and ethical analysis of AI-assisted decision support). The successful candidate will play a central role in shaping the technical architecture, leading the development of a multilingual, federated, explainable generative AI agent that captures patient narratives across physical, psychological, social, and financial burden domains and translates them into guideline-linked, clinically actionable recommendations. The selected candidate will: - Lead the design and development of the ResPECT generative AI agent, including a modular multi-agent architecture (clinical dialogue agent, retrieval/decision agent, guardrail agent, SOAP-note agent) following the g-AMIE paradigm with instruction- tuning, self-play simulated dialogues, and inference-time progressive reasoning. - Adapt and fine-tune open-source European LLMs (e.g., Mistral, EuroLLM) for multilingual clinical narrative classification across burden domains, with intensity scoring, natural-language summaries, and calibrated confidence values. - Build a hybrid retrieval-augmented generation pipeline (RAG-N for narrative classification, RAG-G for guideline-grounded recommendations) combining dense (FAISS) and sparse (BM25) retrieval, multilingual embeddings, and reranking. - Design and implement the Knowledge Fusion Engine integrating PROMs, CTCAE events, lab values, and imaging descriptors via FHIR-compliant schemas to ground generation in clinical context. - Develop explainability-by-design mechanisms (saliency maps, surrogate decision paths, evidence tracebacks), continuous safety monitoring (hallucination detection, bias and drift audits), and guardrail policies aligned with the EU AI Act. - Deploy and orchestrate federated learning across European clinical sites (Flower, PySyft) to enable GDPR-compliant decentralised training while preserving data sovereignty. - Develop the physician voice cloning module (VITS / YourTTS / F5-TTS) and emotion- and literacy-adaptive interaction modules for inclusive, trust-enhancing patient interactions. - Expose REST/OpenAPI services and HL7 FHIR resources for downstream integration in the Medical Assistant (CANKADO) and Policy Assistant (UCAM / WHO-IARC). - Coordinate technical activities with international consortium partners (TABC, CANKADO, Charité, KSE, UCAM, EORTC, NKI) and contribute to publications, conference presentations, and EU reporting. - Co-supervise junior researchers and PhD students working on related topics within the SCOPE Lab and the Digital Surgery & Breast Imaging Lab. Requirements Research Field: Computer science Education Level: PhD or equivalent Skills / Qualifications Minimum Qualifications - 1st degree in Computer Science, Electrical and Computer Engineering, Biomedical Engineering, Applied Mathematics, or a related field. - PhD in Computer Science, Artificial Intelligence, Machine Learning, Natural Language Processing, or a closely related field, with a thesis focused on Large Language Models, agentic AI systems, medical NLP, or clinical decision support. Essential skills - Demonstrated research excellence through first-author/corresponding-author publications in top-tier venues - either machine learning conferences (NeurIPS, ICML, ICLR, ACL, EMNLP, NAACL, AAAI, MICCAI, MIDL) or high-impact medical AI journals (Q1 top 5/10 and Q2). - Strong programming skills in Python and the modern ML ecosystem (PyTorch, HuggingFace Transformers, PEFT, DeepSpeed / FSDP). - Experience with multilingual NLP and adaptation of LLMs to specialised domains, ideally including healthcare. - Familiarity with medical AI evaluation methodology - SPIRIT-AI, CONSORT-AI, TRIPOD-AI reporting standards; OSCE-style human evaluation; psychometric concepts relevant to PROM validation. - Working knowledge of EU regulatory frameworks for medical AI - EU AI Act (high-risk medical AI), GDPR, MDR; awareness of ISO 14971, ISO 27001, ISO 13485, IEC 62304. - Excellent collaborative and communication skills, with proven ability to work in interdisciplinary, multi-institutional teams. Specific Requirements Hands-on experience with modern LLM toolin The candidate is expected to demonstrate hands-on experience across the following tool categories. Equivalent tooling experience accepted where the candidate demonstrates depth in the underlying concepts: - Local inference and serving: vLLM, SGLang, Ollama - Fine-tuning: Unsloth, TRL, Axolotl (LoRA / QLoRA, SFT, DPO / GRPO) - Agent frameworks: LangGraph, DSPy, smolagents - RAG infrastructure: FAISS + BM25, Qdrant, pgvector - Federated learning: Flower, PySyft, or NVIDIA FLARE - Evaluation and observability: LangFuse, Ragas, Inspect - Guardrails and safety: NeMo Guardrails, Giskard - Constrained generation and structured outputs: Outlines, Instructor, XGrammar - Speech and voice: Parakeet, Whisper / Faster-Whisper, XTTS / F5-TTS - Healthcare-specific tools: HAPI FHIR, MedSpaCy, Docling Desirable skills that will also be considered - Prior experience leading or contributing substantially to large multi-partner research projects (Horizon Europe, NIH, or equivalent). - Experience with HL7 FHIR, openEHR, or other clinical interoperability standards in production settings. - Experience with causal inference, counterfactual generation, or generalisability methods in medical AI. - Open-source contributions to relevant libraries (HuggingFace, LangGraph, Flower, vLLM, or similar). - Prior experience co-supervising PhD students or junior researchers. - Working knowledge of Portuguese (not required; English is the working language of the lab and the consortium). Languages ENGLISH Level: Excellent Benefits - Competitive remuneration package commensurate with skills, qualifications, and experience. - Full immersion into a research excellence ecosystem with highly motivated researchers, supported by state-of-the-art technology and continuous development opportunities. - Access to CF`s high-performance computing infrastructure and clinical datasets from the Champalimaud Clinical Centre. - Strong international collaboration with leading partners across 10+ European countries, including direct interaction with the two landmark Lancet Commissions on Cancer and Health Systems and on the Mental Health Effects of the COVID-19 Pandemic. - Substantial opportunities for high-impact publications at top ML conferences and medical journals, conference travel, and visibility within the European generative AI healthcare landscape. - Duration of the Fellowship: 24 months, automatically extended by 12 months s

O que procuramos

• Offer Description
• The Champalimaud Foundation (Fundação D. Anna de Sommer Champalimaud e Dr. Carlos Montez Champalimaud), a private, non-profit research institution in Lisbon, Portugal, is looking for a Postdoctoral Researcher to join the Surgery, Care, Outcomes, Personalization, and Empowerment (SCOPE) Lab, within the Breast Cancer Research Programme.
• The SCOPE Lab is led by Professor Maria João Cardoso (Head of the Breast Unit, Champalimaud Foundation). The postdoctoral researcher will be assisted in their scientific and technical work by Dr. João Santinha (PI and Co-Lead of the Digital Surgery Lab and member of the Breast Imaging Group) and Dr. Luís Elvas (Health Data Engineer of the Breast Cancer Research Programme).
• The selected candidate will lead Champalimaud Foundation´s technical contributions to ResPECT (Representing People´s Experience of Cancer and its Treatment) - one Pan-European generative AI project in healthcare, funded with €16 million by the Horizon Europe programme (HORIZON-HLTH-2025-01-CARE-01). ResPECT brings together leading partners from over 10 countries to develop the first end-user–driven, trustworthy generative AI Virtual Assistant for patients, clinicians, and health systems in oncology and mental health care.

Responsabilidades

• Offer Description
• The Champalimaud Foundation (Fundação D. Anna de Sommer Champalimaud e Dr. Carlos Montez Champalimaud), a private, non-profit research institution in Lisbon, Portugal, is looking for a Postdoctoral Researcher to join the Surgery, Care, Outcomes, Personalization, and Empowerment (SCOPE) Lab, within the Breast Cancer Research Programme.
• The SCOPE Lab is led by Professor Maria João Cardoso (Head of the Breast Unit, Champalimaud Foundation). The postdoctoral researcher will be assisted in their scientific and technical work by Dr. João Santinha (PI and Co-Lead of the Digital Surgery Lab and member of the Breast Imaging Group) and Dr. Luís Elvas (Health Data Engineer of the Breast Cancer Research Programme).
• The selected candidate will lead Champalimaud Foundation´s technical contributions to ResPECT (Representing People´s Experience of Cancer and its Treatment) - one Pan-European generative AI project in healthcare, funded with €16 million by the Horizon Europe programme (HORIZON-HLTH-2025-01-CARE-01). ResPECT brings together leading partners from over 10 countries to develop the first end-user–driven, trustworthy generative AI Virtual Assistant for patients, clinicians, and health systems in oncology and mental health care.
• CF leads Work Package 3 (Generative AI agent and system architecture) of the project, with key responsibilities also across WP4 (Medical and Policy Assistants), WP5 (LLM output validation), and WP7 (legal and ethical analysis of AI-assisted decision support). The successful candidate will play a central role in shaping the technical architecture, leading the development of a multilingual, federated, explainable generative AI agent that captures patient narratives across physical, psychological, social, and financial burden domains and translates them into guideline-linked, clinically actionable recommendations.