02
02 — Case Studies
Delivered at Scale.
Cloud Architecture
Two production-style AWS builds: one focused on fault tolerance under load, the other on independent deployability and cost efficiency.
FEATURED CASE STUDY
Cloud-Native Microservices & CI/CD Platform
Problem
A monolithic application needed to be decomposed into independently deployable services with automated delivery pipelines.
Containerized services deployed to Amazon ECS with Fargate using ARM64 Graviton for cost-optimized serverless compute. Implemented fully automated CI/CD via AWS CodePipeline and CodeDeploy with a blue/green deployment strategy, backed by Amazon RDS MySQL in private subnets.
$87
Monthly Infra Cost
0-DT
Zero-Downtime Deploys
100%
Fault Isolation / Service
CloudAWS
Fault-Tolerant AWS Admissions Architecture
Designed and deployed a fault-tolerant architecture for a university admissions application across two Availability Zones. EC2 instances run in private subnets behind an Application Load Balancer, with Auto Scaling validated to four instances under simulated peak traffic and Amazon RDS Multi-AZ synchronous replication providing database high availability.
2 AZ
Fault Tolerant
4x
Auto Scaling Validated
Multi-AZ
RDS Replication
AWSRegulated Industry
Pipeline Risk Modeling Platforms (Oil & Gas)
Directed multimillion-dollar, AWS-hosted SaaS implementations supporting probability-based risk modeling across transmission, distribution, and storage pipeline assets for major U.S. pipeline operators. Led cross-functional squads of 8 through requirements, UAT, and production deployment in full compliance with PHMSA, the federal framework governing U.S. hazardous liquid and gas pipeline operations.
⚡Transmission ›
🏠Distribution ›
🏭Storage ›
PHMSA
Regulatory Compliance
8
Cross-Functional Squad Size
IntegrationsJira
LMS & Data Integration Programs
Led a strategic vendor integration program on the Canvas LMS platform, used across hundreds of universities and colleges worldwide, coordinating directly with third-party vendors to deliver integrations (Turnitin, Power BI, SMS) aligned to institutional priorities and roadmap planning, managed and tracked through Jira.
🎓Canvas LMS ›
🔗Vendor Integrations ›
3
Vendor Integrations Shipped
Global
Platform Scale (Canvas LMS)
Cloud Security Automation
Simulated Attack Chain Case Studies
Three simulated AWS attack chains: privilege escalation via Lambda, privilege escalation via EC2 instance profiles, and a full SSRF to breach simulation modeled on the 2019 Capital One incident. The common thread is dangerous permission combinations, not single misconfigurations, being what actually gets exploited.
FEATURED CASE STUDY
S3 Data Exfiltration Breach Simulation
Problem
Demonstrate how a real-world attack pattern chains SSRF, exposed instance metadata, and an over-permissioned IAM role into a full data breach.
Provisioned a vulnerable AWS environment with Terraform (17 resources: VPC, EC2 reverse proxy, IAM role, S3 bucket) modeled on the 2019 Capital One incident. Exploited a Host-header SSRF vulnerability to reach the EC2 Instance Metadata Service, extracted temporary IAM credentials, and used them to enumerate and exfiltrate data from a restricted S3 bucket. Documented full root-cause analysis and defensive recommendations, including enforcing IMDSv2.
17
AWS Resources Provisioned
1
Named Breach Recreated
Full
Teardown Post-Lab
IAMLambda
Privilege Escalation via Lambda
Starting from a read-only IAM user, chained
sts:AssumeRole, Lambda management permissions, and iam:PassRole to reach an overprivileged Lambda execution role. Deployed a payload that attached AdministratorAccess to the original low-privilege user, fully escalating access.Read-Only
Starting Access
Admin
Ending Access
IAMEC2
Privilege Escalation by Attachment
Exploited a combination of
iam:PassRole, ec2:RunInstances, and iam:AddRoleToInstanceProfile to swap a low-privilege role for an admin role on an EC2 instance profile. Stole temporary administrator credentials from the Instance Metadata Service and used them to terminate a protected EC2 instance, proving real impact.IMDS
Credential Theft Path
Verified
Impact Proven
MFA / SSOPayments
Secure Portal & Authentication Enhancements
Implemented Multi-Factor Authentication and Single Sign-On across enterprise portals, and led the integration of a secure payment API with CIBC, one of Canada's largest Schedule I chartered banks, enabling students to transact directly within institutional platforms. Production security work, not a simulated environment.
🔐MFA & SSO ›
🏢CIBC Payment API ›
MFA+SSO
Enterprise Auth Hardening
CIBC
Bank-Grade Payment API
Agentic AI & Workflows
Applied Agentic AI Case Studies
FEATURED CASE STUDY
RAG Document QA System
Problem
Document question-answering needs grounded, source-cited answers, not fluent-sounding guesses.
Built a 15-document research corpus, indexed embeddings in Chroma, and developed a
search_knowledge_base retrieval tool. Added multi-query rewriting to improve retrieval coverage and citation-grounded generation to reduce hallucination. Evaluated with an LLM-as-judge faithfulness evaluator.1.00
Avg. Faithfulness Score
15
Document Corpus
100%
Cited Generation
FEATURED CASE STUDY
First AI Chatbot Implementation
Problem
Students and applicants across 7+ institutions had no 24/7 self-service support, placing unsustainable load on staff and creating inconsistent information experiences.
Led the organization's first AI implementation from strategic inception through scaled multi-institution deployment, coordinating directly with the Executive Director of Digital Transformation, the Executive Director of IT & Infrastructure, and the VP of Operations. Conducted a structured build vs. buy evaluation, proceeded with a third-party proof of concept, then managed the full implementation lifecycle including cross-departmental knowledge base development. Established a dual monitoring framework pairing a dedicated accuracy review team with hallucination safeguards.
7+
Institutions Deployed
24/7
Availability
Dual
Accuracy Monitoring Layer
AgentsMulti-Model
AI Agent Enhancement Project
Expanded an existing AI agent from a basic prototype into a more product-oriented application. Added multi-model testing across OpenAI and Claude to compare response quality, performance, and operating cost, and scoped a usage dashboard covering token consumption, latency, and estimated spend.
2
Providers Compared
Cost
Aware Reporting
MLXGBoost
Imbalanced-Data Fraud Detection
Built a machine learning pipeline for credit card fraud detection on a highly imbalanced dataset, where fraud represented roughly 0.17% of transactions. Compared logistic regression and XGBoost across multiple imbalance-handling techniques, with threshold tuning on the precision-recall curve delivering the strongest practical tradeoff between catching fraud and limiting false alerts.
0.17%
Fraud Rate in Dataset
XGBoost
Strongest Tradeoff