Machine Learning Resident - Client: Theragraph; term

Position: Machine Learning Resident - Client: Theragraph (6 month term)
“If you are excited by the challenge of using ML to make sense of large, complex and disparate health datasets, this is the right opportunity for you. Be a part of a collaborative team of clinicians, scientists and engineers building a real world solution that aims to bring clarity and efficiency to clinical practice.”

Dave Staszak, Lead Machine Learning Scientist, Advanced Technology

Description

About the Role:

This is a paid residency that will be undertaken over a 6-month period with the potential to be hired by our client, Theragraph, afterwards (note: at the discretion of the client). The Resident will report to an Amii Scientist and regularly consult with the client team to share insights and engage in knowledge transfer activities. Successful candidates will be members of a cross‑functional project team with backgrounds in ML research, project management, software engineering, and new product development.

This is a rare opportunity to be mentored by world‑class scientists and to develop something truly impactful.

About The Client
Theragraph is an early stage digital health startup based in Edmonton, Alberta.

We simplify the complexity of health records to help specialist physicians make better decisions in less time. At the same time, we provide patients with never‑before accessible visibility into their disease history and pharmaceutical companies with the data they need to innovate. Founded and led by practicing physicians, Theragraph is built to be overwhelmingly useful for healthcare providers and a bridge for harnessing health data towards better health outcomes.

We are a small, scrappy team of intelligent, pro‑active and curious people. We are currently fully remote but like to gather in person when we can.

About The Project
This project involves the development of an intelligent data extraction pipeline from unstructured data source to structure complex medical records for Crohn’s and Ulcerative Colitis patients. We utilize an open‑source 20B parameter GPT model to process unstructured OCR text into high‑fidelity clinical datasets, supported by a robust two‑stage Human‑in‑the‑Loop (HITL) framework. This system captures real‑time expert corrections for both noisy machine‑generated text and LLM structural errors, generating a rich repository of preference pairs and audit logs.

The primary objective is to apply advanced Reinforcement Learning techniques (such as RLHF or GRPO) to fine‑tune the LLM, enabling the model to learn directly from clinical subject‑matter experts’ evaluations. Built on a scalable AWS architecture, the project focuses on minimizing model hallucinations and managing the “alignment tax” to ensure precise medical reasoning and formatting. By closing the feedback loop between human validation and model training, we aim to achieve production‑grade accuracy in the automated capture of specialized health data.

Required Skills / Expertise
Are you passionate about building great solutions? You’ll be presented with opportunities to both personally and professionally develop as you build your career. We’re looking for a talented and enthusiastic individual with a solid background in machine learning, specifically human‑in‑the‑loop reinforcement learning techniques, ideally with some experience with unstructured data extraction methods (OCR‑, LLM‑based, etc.).

Key Responsibilities

Design, implement, optimize, and evaluate models to accurately capture health data across a range of data types, formats, and domains.

Apply advanced reinforcement learning methods (e.g., RLHF, GRPO) and subject matter expert evaluations to learn from mistakes and improve data collection processes.

Conduct applied research on reinforcement learning, OCR and RAG techniques to overcome limitations in current approaches.

Optimize ML pipelines to ensure efficiency, scalability, and real‑time processing capabilities.

Implement specific metrics to measure “alignment tax” and ensure that RLHF improvements in accuracy do not degrade the model’s general reasoning or formatting capabilities.

Collaborate with the project team and stakeholders to develop MVP and client focused solutions.

Engage in regular client…