Overview
Are you passionate about large language models (LLMs) and healthcare? Do you want to translate cutting-edge NLP into tools that clinicians can safely use at the point of care? We’re hiring a full-time Technical Research Assistant to help design, build, and evaluate LLM-powered clinical decision support systems.
This role is ideal for graduate students, recent grads, or practicum/capstone students who want hands-on experience applying LLMs to clinical workflows. You’ll work with clinical text and knowledge sources (guidelines, reports, EHR notes) and contribute directly to retrieval-augmented generation (RAG), fine-tuning, and rigorous evaluation of models intended for real-world clinical impact.

What You’ll Do

• Build RAG pipelines that ground LLM outputs in trusted clinical sources (guidelines, formularies, structured knowledge bases).
• Curate and preprocess datasets from clinical text and semi-structured data (PHI de-identification, normalization, ontology mapping to SNOMED CT / ICD-10 / LOINC).
• Design prompts, tools, and function-calling schemas for CDS tasks (summarization, guideline lookup, order suggestion, discharge instructions).
• Train and evaluate models (instruction-tuning, preference optimization) using frameworks like PyTorch/Transformers and orchestration libraries (e.g., LangChain/LlamaIndex).
• Implement vector search and indexing (FAISS/Milvus/pgvector) and optimize retrieval quality (chunking, hybrid search, re-ranking).
• Stand up robust evaluation: automatic metrics (exact match/F1, citation accuracy, groundedness) and human-in-the-loop reviews with clinicians using structured rubrics.
• Build guardrails and safety checks (PHI leakage tests, hallucination detection, contraindication checks, prompt-injection resilience).
• Maintain reproducible, well-documented code and experiments (Git, Jupyter, Docker; experiment tracking).
• Collaborate closely with clinicians, data stewards, and AI researchers; participate in study design, IRB/REB documentation support, and write-ups for publications.

Why Join Us

• Gain hands-on experience with one of the most promising areas of healthcare AI: pediatric neuroimaging
• Eligible to be counted as practicum or experiential learning credit for graduate or professional programs (check with your program advisor)
• Learn to develop, deploy, and validate AI models in a healthcare research environment
• Receive mentorship from clinicians, AI researchers, and imaging scientists
• Contribute to peer-reviewed publications and high-impact research
• Work with large-scale imaging data from leading hospitals and research centers

Required Skills & Background

• Bachelor’s degree in Computer Science, Biomedical Engineering, Physics, or related field (Graduate students encouraged to apply)
• Strong Python skills and experience with deep learning/LLM stacks (PyTorch; Hugging Face Transformers).
• Familiarity with retrieval and LLM application patterns (RAG, tool use/function calling, prompt engineering).
• Experience building data pipelines and parsing clinical or technical text (regex/NLP, basic SQL).
• Comfort with experiment design and evaluation for QA/summarization tasks (EM/F1, calibration, citation/attribution checks).
• Proficiency with Git, UNIX, and Jupyter; attention to reproducibility and clean code.

Nice-to-Have

• Exposure to healthcare data standards or ontologies (HL7 FHIR, SNOMED CT, ICD-10, LOINC).
• Experience with vector databases, re-rankers, or hybrid search.
• Knowledge of privacy, security, and compliance in Canadian contexts (PHIPA/PIPEDA) and general best practices for handling sensitive data.
• Prior work on safety/guardrails, red-teaming, or evaluation frameworks (e.g., rubric-based human evals, RAGAS checks).

Application Requirements

• Resume/CV
• Brief cover letter outlining technical experience and interest in the position
• Unofficial transcript
• GitHub portfolio, code samples, or publications (optional but encouraged)

Job Type: Full-time
Pay: $30.00-$40.00 per hour
Expected hours: 37.5 per week
Work Location: Hybrid remote in Toronto, ON M5S 1A

• Build RAG pipelines that ground LLM outputs in trusted clinical sources (guidelines, formularies, structured knowledge bases).
• Curate and preprocess datasets from clinical text and semi-structured data (PHI de-identification, normalization, ontology mapping to SNOMED CT / ICD-10 / LOINC).
• Design prompts, tools, and function-calling schemas for CDS tasks (summarization, guideline lookup, order suggestion, discharge instructions).
• Train and evaluate models (instruction-tuning, preference optimization) using frameworks like PyTorch/Transformers and orchestration libraries (e.g., LangChain/LlamaIndex).
• Implement vector search and indexing (FAISS/Milvus/pgvector) and optimize retrieval quality (chunking, hybrid search, re-ranking).
• Stand up robust evaluation: automatic metrics (exact match/F1, citation accuracy, groundedness) and human-in-the-loop reviews with clinicians using structured rubrics.
• Build guardrails and safety checks (PHI leakage tests, hallucination detection, contraindication checks, prompt-injection resilience).
• Maintain reproducible, well-documented code and experiments (Git, Jupyter, Docker; experiment tracking).
• Collaborate closely with clinicians, data stewards, and AI researchers; participate in study design, IRB/REB documentation support, and write-ups for publications