Generative AI Engineer, Healthcare

About the Role

Recollia builds AI for medical students, including tutors, clinical cases, exam questions, and image Q&A, all generated with AI and verified by doctors. This role focuses on the generative AI aspects of these products.

WHERE THIS GOES

This position begins as an internship with a stipend of Rs 20,000 and a potential bi-weekly incentive of Rs 5,000, based on performance. The goal is to transition into a founding engineer role with equity upon demonstrating the ability to produce generated content (images, cases, questions) that doctors approve and students learn from.

Key Responsibilities

• Medical Image Generation:
• Create anatomy, histology, pathology, and radiology-style teaching images, as well as mechanism diagrams.
• Fine-tune diffusion models to ensure anatomical accuracy and manage cost per image at scale.
• Clinical Cases and MCQs:
• Generate exam-grade clinical vignettes and multiple-choice questions (MCQs) aligned with standards like NEET-PG and USMLE.
• Ensure content is grounded in sources, structured, and reviewed before student use.
• Image Q&A:
• Develop a multimodal tutor capable of analyzing medical images (X-rays, ECGs, slides) and providing educational insights.
• Corpus Development:
• Build and maintain a versioned, licensed, and de-identified medical education dataset comprising images, text, and questions.
• Evaluation and Feedback Loops:
• Implement evaluation systems and review processes involving doctors to ensure medical accuracy.
• Integrate clinician feedback to refine models.

Requirements & Tech Stack

• Generative Images:
• Proficiency with diffusion models, including fine-tuning techniques (LoRA, DreamBooth or similar), ControlNet, and inpainting.
• Ability to maintain consistent style across numerous images.
• Understanding of anatomical accuracy as a critical constraint.
• Experience evaluating generated images beyond metrics like FID, utilizing expert review and failure taxonomies.
• Implementing fairness in generation, considering diverse skin tones, ages, and sexes.
• Generative Text:
• Expertise in generating clinical cases and questions using LLMs, focusing on structured outputs, source grounding with citations, and hallucination control.
• Understanding of item-writing principles for MCQs (fairness, distractor effectiveness, Bloom's levels, NBME-style guidelines).
• Experience measuring question quality with student data (difficulty, discrimination).
• Multimodal:
• Experience with vision-language models (e.g., Qwen-VL, LLaVA, MedGemma) and their fine-tuning on image-text pairs.
• Ability to evaluate performance on medical image Q&A and identify limitations.
• Skills in building image-text datasets, including captioning, labeling, review, and understanding licensing.
• Data and Corpus:
• Experience with dataset pipelines for ingestion, cleaning, deduplication, de-identification, and versioning.
• Familiarity with synthetic data generation and expert review.
• Understanding of licensing and consent for medical content.
• Evaluation and Safety:
• Proficiency in medical accuracy evaluations with clinicians, LLM-as-judge methodologies, and regression testing.
• Ensuring all generated images are clearly labeled as synthetic and educational.
• Medicine:
• Genuine curiosity about medical education.
• Willingness to collaborate closely with doctors and students to understand medical learning processes and identify model inaccuracies.
• Engineering Foundations:
• Production-level proficiency in Python and PyTorch, including libraries like diffusers and transformers.
• Experience with async pipelines and writing tests.
• Experience with GPU computing, including fine-tuning on platforms like RunPod, Modal, or AWS.
• Knowledge of quantization, memory optimization, and cost-per-asset calculations.
• Familiarity with FastAPI or equivalent, Postgres, S3, Docker, Git, and CI/CD.
• Experience with experiment logging tools like W&B or equivalent.

Qualifications

• No degree is required; skills and experience are prioritized.
• Demonstrated experience in generating a significant volume of AI outputs (e.g., images) and critically evaluating them.
• A proactive approach to identifying and rectifying errors.
• A desire to take on broader responsibilities and contribute to decision-making.
• The ability to provide constructive criticism and identify issues, including with company decisions.

How to Apply

Email careers@recollia.ai with the subject "Generative AI, Healthcare". Include:

• One project built with generative AI (image, text, or multimodal) with a link to a repo, demo, or write-up.
• In three sentences, describe your project, a challenge encountered, and what you would do differently.
• Generate one medical teaching image using any tool, attach it, and note potential corrections a doctor might suggest.