Andrew Kujtan

Andrew Kujtan

Successful background in designing and maintaining robust and performant software systems incorporating AI functionality backed by programmatic testing and static analysis. Possesses strong intuition in AI system design, program debugging, and fault analysis with a focus on using the right tool for the job.

Work

Scotiabank

Director, AI
- Present

Lead a team in developing AI-based solutions to optimize and automate existing workflows, driving improved efficiency and quality control across processes. Works closely with stakeholders to ideate and develop solutions that meet business needs while ensuring compliance with regulatory requirements.

  • Led the development and production launch of an AI-powered mortgage review platform that automated document extraction, compliance validation, and exception detection across every mortgage application, achieving 30% faster review times.

  • Implemented an automated contract review system that effectively identified language and financial discrepancies. System reviews over $500 billion in contracts annually.

  • Engineered a document verification system to fully automate mortgage insurance regulatory obligations, bringing coverage from 2% to 100%.

  • Launched an email triage system that automated sorting processes of 30K emails/month, cutting down triage times from hours to minutes. Utilizing a fine-tuned BERT derivative for classification and Gemini for information extraction. Achieved 95% accuracy compared to human performance of 85%.

  • Founded and led LLM/Generative AI communities of practice to drive innovation and cross-team collaboration.

  • Lead development of multiple ML/LLM based applications automating document processing workflows, including mortgage, insurance, and contract review processes, resulting in significant efficiency gains and improved accuracy.

  • Collaborate with cross-functional teams to integrate vendor systems using AI/ML models, setting and validating performance metrics that ensured services met contractual requirements.

  • Manage a team of 8+ data scientists and engineers, providing mentorship and guidance to foster professional growth.

Scotiabank

Senior Manager, Data Science
-

Spearheaded the development of internal applications that accelerated partner activities and streamlined data analytics processes across the bank.

  • Developed graph traversal tooling that provided in-depth insights for marketing, wealth, and fraud teams, leading to more targeted strategies.

  • Engineered robust data pipelines to support the creation of a scalable data lake, leveraging Spark and SQL for efficient data ingestion and transformation.

  • Mentored interns and junior team members, sharing industry best practices and fostering professional development.

Evertz Microsystems Ltd.

Design Engineer
-

Developed a native web client for controlling a broadcast network utilized by leading North American television providers.

  • Developed a single-page web application using a Spring MVC backend (Tomcat/Jetty) and a Backbone/jQuery frontend, efficiently handling large datasets in high-latency environments.

  • Automated the manual build process by tracking source changes, reducing turnaround time from weeks to minutes, and implemented static analysis tools to proactively catch errors.

  • Successfully transitioned over 10 million lines of code from Java 6 to Java 8.

Education

Publications

System and Method for Reviewing and Evaluating Discrepancies Between Two or More Documents

Published by USPTO, Dec 2023

The present disclosure pertains to a system and method for determining differences between at least two documents. A preferred embodiment disclosed herein pertains to a system and method for determining differences in the text or terms of legal documents, for example transactional legal agreements, such as but not limited to Trade Confirmations. The document review system can compare text documents to other text documents but is further capable of comparing text documents to structured spreadsheet data. The combination of static component verification in combination with dynamic component verification leads to a powerful tool which greatly reduces the time and efficiency of document review while reducing the risk of human errors. Utilizing word embeddings in the static verification in combination with the use of named entity recognition, rule parsers in the dynamic verification, enables the system to produce concise and meaningful summaries of differences between multiple documents.

Skills

ML / Deep Learning

Master
  • transformers
  • Machine Learning
  • Information Extraction
  • Model Evaluation
  • Gemini
  • Vertex AI

Data Science

Master
  • python
  • pandas
  • numpy
  • polars
  • Power BI

Data Engineering

Master
  • Git
  • Spark
  • SQL
  • python
  • Docker
  • Google Cloud Platform

Project Management

Master
  • Agile
  • Scrum
  • Kanban
  • Jira
  • Confluence

Web Development

Proficient
  • HTML
  • CSS
  • JavaScript

Languages

  • English - Native speaker

Interests

Recreational Sports

  • Hockey
  • Volleyball
  • Cross-Country Ski

Hobbies

  • Beer Making
  • Bread Making
  • Hiking