About Me
Engineering AI & Data Systems at Scale
I am Qijin Xu (Jack), a Toronto-based lead AI and machine learning engineer delivering production AI systems that combine robust data engineering with modern LLM agents to move critical business metrics.
Bio
Summary
Lead AI/ML Engineer delivering measurable impact via production AI systems. Drove 90% support reduction (525% ROI) with RAG for 50k+ users, and 50% labor reduction with LLM agents (LangGraph), drawing on full-stack expertise from data systems (Spark, Kafka), classic ML (321% fraud lift) to modern LLMs.
Expertise & Stack
Python, Java, Go, TypeScript, SQL, Azure, AWS, GCP, Spark, Kafka, Airflow, LangGraph, RAG, MCP, OpenAI API, and full-stack ownership from ingestion to production AI services.
Credentials
B.Sc. in Computer Science, York University | Microsoft Certified: Azure Solutions Architect Expert | Project Management Professional (PMP).
Beyond Code
Open-source contributor, avid gamer, and builder of assistant tools that blend computer vision, machine learning, and gameplay analytics.
Skills
AI & LLM Engineering
Large Language Models (LLMs)
LangGraph agent workflows
Retrieval-Augmented Generation (RAG)
Model Context Protocol (MCP)
OpenAI API / vLLM / Hugging Face
Prompt and structured output design
LoRA / PEFT fine-tuning
Backend & System Design
Distributed systems and scalability
Microservices and service mesh
API design (REST / GraphQL)
Database design (SQL and NoSQL)
Event-driven architecture and CQRS
Message queues (Kafka, RabbitMQ)
Cloud & DevOps
Azure (Functions, AKS)
AWS (EKS, Lambda, S3, ECS)
GCP (Vertex, BigQuery)
Docker and Kubernetes
Terraform and infrastructure as code
CI/CD (GitHub Actions, Azure DevOps)
Observability (CloudWatch, Prometheus)
Programming Languages
Python
Java
Go
JavaScript / TypeScript
SQL (T-SQL, PostgreSQL)
C++
Bash / Shell scripting
Data & ML Infrastructure
Apache Spark
Kafka
Airflow
Databricks and Delta Lake
OpenLineage data lineage
Vector stores (Pinecone, FAISS, Weaviate)
Redis
Feature stores (Feast)
Full-Stack Development
React / Next.js
tRPC and GraphQL
Tailwind CSS
Vercel and edge functions
Auth0 and OAuth2
Testing (Jest, Playwright)
Experience
CGI Inc.
Lead AI/ML Engineer (Tech Lead), AI Center of Excellence
LangGraphLLM AgentsRAGvLLMPII Filtering
Jan 2025—Present
- LangGraph dual-state agent for a proprietary coding language: orchestrated vLLM-served LLMs with retrieval, EBNF structured outputs, and PII filtering to automate documentation, business-rule extraction, and migration readiness; targeting a 50% labor reduction for support teams.
Data Architect and Engineering Lead, Senior Consultant
LangChainRAGAzureAWSGCPOpenLineageAxiomSL
Aug 2023—Dec 2024
- CGI Impact Award 2024: Architected and shipped an LLM-powered enterprise RAG chatbot (LangChain across Azure, AWS, GCP) delivering about 90% autonomous resolution for a 50k+ user base and 525% ROI in 12 months.
- Built an OpenLineage-based lineage tool that auto-documents extensible multi-source flows with LLM-assisted root-cause suggestions, shrinking investigation cycles across three incidents per month.
- Automated AxiomSL rule updates with an LLM copilot for RBC, cutting manual updates by 80% while preserving governance controls.
Data Engineer, Consultant
PythonSQL (T-SQL, PostgreSQL, SQL Server)Big DataMachine LearningData Warehousing / ETL
Aug 2021—Jul 2023
- Managed the largest automobile-insurance policy-history database in Canada (2 TB on SQL Server) with 99.9% uptime and median queries under 200 ms.
- Delivered production ETL and ELT pipelines for 22+ micro data marts (each under six tables with monthly SLAs), trimming report runtimes from 30 minutes to 5 minutes.
- Shipped a Gradient-Boost plus Logistic Regression loss-ratio model that recovered fraud losses 321% versus a top-5 bank benchmark through a Python and T-SQL pipeline.
Elder Laboratory, York University
Research Assistant, Computer Vision (Part-Time)
Computer VisionPyTorchTensorFlowCNN-LSTMActive LearningReal-Time Systems
May 2020—Aug 2021
- Developed AttentiveVision, a computer-vision pipeline for hockey broadcast video in Python (PyTorch, TensorFlow, OpenCV), training a CNN-LSTM stack with mean average precision above 70% on event detection across 18 hours of game footage.
- Engineered a real-time player tracking system using a Kalman-filter tracker that delivered under 100 ms latency at 30 FPS, capturing pose and movement for live speed and heat-map analytics.
- Created an event-detection annotation loop (goal, timeout, penalty) with active learning that cut manual labeling effort and raised dataset reliability to Cohen's kappa of 0.92.
Nascent Digital
Full-Stack Developer Intern
ReactTypeScriptCI/CDPuppeteerJest
May 2019—Aug 2019
- Built end-to-end features with React, TypeScript, and Node.js; contributed major components to company-wide homepage rewrite.
- Shipped TELUS product pages by partnering with UX/QA/DevOps and creating reusable React components plus CI/CD hooks; finished design-to-deploy in 7 days.
- Introduced automated visual-regression testing with Nightwatch.js and Puppeteer; removed ~90% of manual QA checks.
Education
B.Sc. in Computer Science (Specialized Honours)
York University, Toronto
2017—2021
- Graduated with GPA 3.9/4.0
- First Class with Distinction
Certifications
Microsoft Certified: Azure Solutions Architect Expert
Microsoft
Issued: 2022
Google Cloud Professional Cloud Developer
Issued: 2023