Jacques Delfrate.

SENIOR AI ENGINEER / MONTRÉAL

AI engineering,
from training
to production.

I’m Jacques. I train, optimize, deploy, and monitor AI systems across images, video, signals, and documents.

Explore my workLet’s talk on Upwork

9+ years building AI products · French & English

Training & evaluationOptimization & integrationCloud & edge deploymentMonitoring & reliability

01 / SELECTED WORK

Engineering, with evidence.

Products and platforms I’ve contributed to.
My role, the engineering, and the result.

HEALTHCARE AI / CURRENT WORK

Making imaging AI
part of the workflow.

PACS-AI connects AI models with medical imaging workflows. My work focuses on the engineering around the model: getting studies in, running inference, and making results accessible.

MY CONTRIBUTION

DICOM ingestion, study-completion logic, inference services, deployment infrastructure, result views, and tenant-scoped access. For LLM orchestration within the platform, see my CardioAgent work.

  • Python
  • Docker
  • Orthanc
  • LangGraph
Visit PACS-AIRead deployment preprint

Research and evaluation platform; model use depends on its validation and regulatory status.

LLM AGENTS / CURRENT WORK02

CardioAgent

Connecting language models with clinical AI tools.

I develop a stateful, LangGraph-based agent that coordinates specialized models across ECG, echocardiography, and coronary angiography. Integrated into PACS-AI, it combines retrieval, tool execution, and synthesis using self-hosted LLMs.

MY CONTRIBUTION

Agent development and PACS-AI integration, alongside self-hosted LLM deployment with vLLM, agent tracing with Langfuse, and LLM-as-judge evaluation design.

  • LangGraph
  • ReAct
  • Retrieval
  • vLLM
  • Langfuse
  • LLM evaluation

RETRIEVE. EXECUTE. SYNTHESIZE.

  1. 01
    Retrieve context

    Bring relevant information into the workflow.

  2. 02
    Coordinate tools

    Call specialized models for signals and imaging.

  3. 03
    Synthesize results

    Bring model outputs together in PACS-AI.

Evaluation & tracing across the workflow

Illustrative research workflow.

CONNECTED SYSTEMS03

DeepECG

Reliable infrastructure for ECG AI.

I work on backend deployment pipelines and connected-device infrastructure for DeepECG, with a focus on hardening, secure transmission, and operational reliability.

MY CONTRIBUTION

Hardened backend deployment pipelines. Configured and hardened a Raspberry Pi ECG relay over an AWS VPN, and designed authenticated heartbeats, connectivity checks, outage detection, and recovery notifications.

  • Backend deployment
  • Pipeline hardening
  • Raspberry Pi
  • AWS
Visit DeepECGRead platform paper

Investigational platform; regulatory approvals pending.

COMPUTER VISION / 36PIX04

Background removal at scale.

From segmentation research to a shipped product.

Developed ChromaStarAI background-removal technology used to process approximately 60 million images annually for professional photography.

42%less inference time
26%smaller model
Explore my contribution

Designed the proprietary segmentation model, training pipeline, augmentation, and custom evaluation metrics. Optimized inference with OpenVINO in C++ and Python, and deployed models on AWS EC2 with monitoring and API integration.

  • Segmentation
  • OpenVINO
  • C++
  • AWS
ALSO AT IONODES

Computer vision at the edge.

Real-time license-plate recognition and OCR on Intel NUC, optimized with OpenVINO in C++. Android body-camera face-blurring proof of concept.

2022 — 2023

02 / HOW I CAN HELP

One engineer. The full AI lifecycle.

01

Train & build.

Prepare data, train and fine-tune models, and evaluate performance. Build computer vision pipelines, retrieval systems, and LLM agents around your product needs.

02

Optimize & integrate.

Improve inference speed, model size, and accuracy. Connect AI to APIs, data pipelines, and existing software, with clear performance trade-offs.

03

Deploy & monitor.

Deploy on cloud or edge hardware, harden backend deployment pipelines, and add observability, connectivity checks, and recovery notifications.

03 / SELECTED PUBLICATIONS

Research behind the work.

Selected work listing me as a co-author.
Journal articles and preprints, clearly identified.

2026
arXiv Preprint

Lessons learned from deploying imaging AI with the open PACS-AI platform

Integration, feedback, and infrastructure for imaging AI deployment.

Read paper 2026
medRxiv Preprint

A unified 12-lead ECG-language model for interpretation and clinical-endpoint prediction

DeepECG-Tok: connecting ECG representations with language models.

Read paper 2026
European Heart Journal Journal article

Foundation models for electrocardiogram interpretation: clinical implications

Supervised and self-supervised ECG models and external validation.

Read paper 2026
Nature Cardiovascular Research Journal article

Multiview deep learning improves detection of major cardiac conditions from echocardiography

Combining information from multiple echocardiographic video views.

Read paper
More publications & conference abstracts

04 / ABOUT

Research depth.
Product experience.

I’m an AI engineer and technical mentor based in Montréal, with 9+ years of experience turning research and product requirements into working software.

My background is in biomedical and electrical engineering. I work with both technical and clinical teams, from architecture and implementation to validation and handoff.

CORE TOOLKIT

Python · C++ · PyTorch · OpenCV · LangGraph · vLLM · Docker · AWS

Montréal, Canada / French & English

LET’S BUILD SOMETHING USEFUL

What’s your AI challenge?

Tell me what you’re building, what already works,
and where you need help.

Let’s talk on Upwork