Devanshu Khurma

AI/ML Engineer
Solutions Architect
Forward Deployed Data Scientist
ML Ops Engineer

Applied Machine Learning Scientist II at TD Bank Group and Computer Science M.S. graduate from the University of Texas at Austin (4.0 GPA). Expert at architecting Agentic AI, custom Generative AI workflows, and cost-efficient cloud systems that deliver documented enterprise savings of over CA$28MM+ annually.

About Me

I am an AI/ML Engineer and Solutions Architect specializing in translating complex business challenges into production-ready, highly scalable machine learning ecosystems. My background sits at the intersection of advanced algorithm development, high-efficiency system design, and strategic business-value generation.

At TD Bank Group, I serve as a primary technical SME, architecting and deploying end-to-end Agentic AI systems utilizing Large Language Models (LLMs) and Vision Language Models (VLMs) on Azure Databricks. My work has reduced underwriting turnaround times from 15 hours to 15 minutes, generating CA$15 million in annual savings and resulting in 4 filed patents.

I hold a Master of Science in Computer Science from the University of Texas at Austin (4.0 GPA) and a Master of Management in Analytics from McGill University (3.9 GPA, Dean's List). This blend of deep computer science knowledge and quantitative business analytics enables me to build technically elegant solutions that scale seamlessly while optimizing compute cost and infrastructure pipelines.

Devanshu Khurma Professional Profile Photo

Technical Expertise

AI, ML & GenAI
PyTorch TensorFlow Hugging Face LangChain LangGraph Agentic AI PEFT (LoRA) Vector Embeddings RAG Systems LLMOps Anomaly Detection
Cloud, Data & DevOps
AWS Azure GCP Databricks Apache Spark Snowflake Apache Kafka Docker Apache Airflow CI/CD Pipelines
Languages & Analytics
Python SQL R Scala Pandas Scikit-learn Tableau Power BI
Certifications & Accolades
AWS Solutions Architect Associate Databricks Data Engineer Associate Vector Institute Agentic AI 2.0 4 Patents Filed

Work History

Toronto Dominion Bank Group

September 2022 – Present | Toronto, ON

Applied Machine Learning Scientist II Current Role

February 2025 – Present

CA$15MM
Annual Cost Savings
4 Patents
Conceptualized & Filed
15 Mins
From 15 Hours Processing

Architected and deployed Agentic AI solutions on Azure Databricks for automated mortgage underwriting, cutting document turnaround processing by 98% and driving CA$15 million in annual savings.

Conceptualized, drafted, and filed 4 distinct machine learning patents within 12 months, pioneering novel architectures in enterprise AI automation.

Fine-tuned a RAG knowledge management system utilizing Hugging Face PEFT (LoRA) and custom instruction-response datasets, significantly optimizing context retrieval speed while minimizing inference-level compute costs.

Spearheaded organization-wide AI literacy by organizing and leading GitHub Copilot Bootcamps and technical workshops, accelerating agile software development across engineering units.

Core Stack: Azure Databricks Agentic AI LangGraph PEFT (LoRA) Hugging Face VLMs RAG

Data Scientist II

September 2022 – February 2025

CA$13.5M
FY26 Value Generated
+115%
Precision Increase
+23%
Recall Boost

Architected a deep learning based anomaly detection engine to identify and mitigate high-risk behavioral trends, generating CA$13.5M in value.

Designed and maintained robust end-to-end Responsible AI pipelines to ensure fairness, interpretability, and compliance with strict federal bank regulations.

Served as the primary technical subject matter expert (SME) to translate complex business requirements into scalable workflow architectures for data engineering teams.

Led key data governance integrations to resolve data inconsistencies following massive cloud migrations, ensuring high compliance and zero business disruption.

Core Stack: PyTorch Scikit-learn Deep Learning Responsible AI Cloud Migration Data Governance

Machine Learning Engineer @ Laps

August 2021 – September 2022 | Montreal, QC

200x
Pipeline Acceleration
28%
Support Workload Cut
CA$33K
Annual Efficiency Gains

Optimized large scale compute infrastructure using GCP BigQuery and AWS SageMaker, achieving a 200x pipeline acceleration and saving CA$18K annually.

Developed and deployed a real-time intent detection chatbot utilizing custom word embeddings, cutting support agents' workload by 28% and boosting client response efficiency.

Consulted on and built a highly concurrent recommender API using AWS Personalize, integrating modern CI/CD DevOps workflows to ensure seamless deployments.

Co-authored architectural pitch decks demonstrating the viability of core machine learning solutions to stakeholders, directly supporting successful funding campaigns.

Core Stack: AWS SageMaker GCP BigQuery Recommender APIs Word Embeddings AWS Personalize DevOps

Data Scientist - Go-to-Market Operations @ Brother International

July 2020 – August 2021 | Montreal, QC

CA$25K
Annual Operations Savings
200+ Hrs
Reclaimed Annually
87%
Forecasting Accuracy

Refactored and optimized complex logistics ETL pipelines in Scala, eliminating redundant queries and saving over 200+ hours in manual tracking operations annually.

Designed and launched a custom time-series forecasting model, achieving an 87% accuracy rating (MAPE) and delivering 15% Forecast Value Added over baseline legacy projections.

Translated historical freight and supply data into rich, interactive business dashboards, enabling Go-to-Market operations teams to refine regional logistics operations.

Core Stack: Scala Apache Spark Time Series ETL Optimization Tableau Logistics Analytics

Business Intelligence Consultant @ Canadian National Institute for the Blind Foundation

June 2020 – August 2020 | Montreal, Canada

12 Hours
Saved Every Month
PowerBI
Anomaly Tracking Dashboard
Python & SQL
Multi-Source Data Pipeline

Identified and defined key performance indicators (KPIs) to evaluate user interaction and success across all of **CNIB's online courses**.

Built a robust **data processing pipeline** in Python and SQL to aggregate and combine raw input from multiple independent database sources.

Developed and launched an interactive **PowerBI dashboard** to efficiently track educational KPIs, spot operational anomalies, and automate reports, saving 12 hours of manual analysis each month.

Core Stack: PowerBI SQL Python KPI Definition Data Integration Process Automation

Analytics Consultant (Pro-Bono) @ Thinkr Consulting

January 2020 – May 2020 | Montreal, Canada

Scorecard
Credit Evaluation Model
Interpret
Point-Based Credit System
Mkono
Kenya Microfinance Support

Conducted **pro-bono consulting for Mkono**, a Kenyan microfinance non-profit organization supporting entrepreneurs with affordable loans and mentorship.

Created and deployed a robust **credit scorecard-based machine learning model** to evaluate microfinance loan viability for local entrepreneurs.

Translated complex, multi-dimensional customer behavioral profiles into a highly interpretable and actionable **point-based credit ranking system** for loan officers.

Core Stack: Credit Scorecard Machine Learning Explainable AI (XAI) Feature Engineering Microfinance Risk

Solutions Architect - Analytics Consulting Project @ Nestle Canada

October 2019 – June 2020 | Toronto/Montreal, Canada

GCP Cloud
End-to-End ML Deployment
Statistical
Data Generation Models
Behavioral
Shopper Pattern Architecture

Designed and implemented a statistical pipeline to **generate highly realistic indicative data**, closely simulating complex real-world retail and supply-chain scenarios.

Created and mapped the comprehensive **process and solution architecture** to identify, segment, and predict consumer behavioral patterns across retail platforms.

Built, validated, and successfully **deployed end-to-end machine learning models on Google Cloud Platform (GCP)** to track shopper trend dynamics.

Core Stack: Google Cloud Platform Solution Architecture Statistical Simulation Machine Learning Consumer Analytics

Software Engineer Intern @ Samsung Research and Development Institute

May 2018 – July 2018 | Bangalore, India

Unsupervised ML
User Activity Segmentation
Agile Scrums
On-Schedule Milestones
Git Version
Code Review Collaboration

Implemented unsupervised machine learning techniques to segment user bases based on activity levels, delivering insights to product design teams.

Utilized Agile methodologies in a fast-paced development ecosystem to achieve key project milestones strictly on schedule.

Gained hands-on expertise with Git version control, actively participating in team code reviews and providing constructive architectural feedback.

Core Stack: Unsupervised ML Python Scikit-learn Agile Methodologies Git Code Review

Education

University of Texas at Austin

Master of Science in Computer Science

2024 – 2026 GPA: 4.0 / 4.0

Focused on advanced machine learning algorithms, deep learning neural networks, Generative AI, scalable cloud computing systems, and computational theory.

McGill University

Master of Management in Analytics

2019 – 2020 GPA: 3.9 / 4.0

Dean’s List. Specialised in machine learning business applications, linear programming, mathematical optimization, advanced statistical forecasting, and database engineering.

Delhi Technological University

Bachelor of Engineering in Computer Science

2015 – 2019 Grade: 86.2%

Completed core coursework in data structures, design and analysis of algorithms, database systems, object-oriented software engineering, and computational mathematics.

Featured Projects

Other Engineering Work

PEFT Knowledge RAG

Fine-tuned transformer models using LoRA on custom instructions to power a high-precision knowledge synthesis workspace for client operations, minimizing compute constraints.

Hugging Face PEFT LoRA RAG

GPT-3 News Aggregator

Developed a news aggregation application leveraging GPT-3 for abstractive text summarization of articles. Implemented automated scraping, content summarization, and responsive delivery.

Python GPT-3 OpenAI API Flask News API React

Regime: Fitness App

Full-stack Android app with a well-structured Firestore database, leaderboards, social feeds, and Kotlin Coroutines/Flow for smooth asynchronous data fetching. Evaluated via unit & integration tests.

Android SDK Kotlin Firebase Firestore LiveData Coroutines

Scalable Recommender API

Consulted on and launched a real-time recommender engine utilizing AWS Personalize, optimized to resolve concurrent user demands under tight service latency requirements.

AWS Personalize SageMaker API Gateway

Scala ETL Logistics Pipeline

Re-engineered legacy warehouse extraction tools into highly efficient, modular Spark operations in Scala, mitigating post-cloud discrepancies and reducing manual runtime tracking.

Scala Apache Spark AWS S3

Interests & Pursuits

Running

Training consistently for full and half marathons, with a personal marathon best of 3:17. I also love trail running, with my largest challenge to date being a grueling 51k trail race.

Running

Digital Photography

Capturing urban landscapes, architecture, and wilderness trails. I enjoy exploring composition, light contrast, and post-processing techniques to document unique visual perspectives.

Digital Photography

Chess

Fascinated by geometric puzzle patterns, structural positional advantages, and endgame calculations. I enjoy playing rapid chess games and working through tactical scenarios online.

Chess

Tennis

Active singles player in amateur community leagues. I enjoy the fast-paced coordination, adjusting ball spin and court angles under match conditions, and continuous technical refinement.

Tennis

Language Learning

Fascinated by semantic structures, phonetics, and local cultural contexts. I hold a JLPT N5 certification and am currently studying towards the N4 level. I am particularly interested in exploring the grammars of English, Sanskrit, and Japanese.

Language Learning

Travel

Exploring historical architectures, trying local food markets, and hiking wilderness trails across national parks. Travel allows me to step back, experience new cultures, and gain fresh perspectives.

Travel

Let's Connect

Whether you are looking to hire a business-impact-driven AI/ML Engineer, an enterprise-level Solutions Architect, or need engineering consultation on advanced Agentic AI deployment, my inbox is always open.

LinkedIn GitHub View Resume