AWS-certified engineer working across serverless data platforms, LLM evaluation and full-stack applications — with a bias toward infrastructure that is reproducible and documented.
Projects spanning cloud data infrastructure, applied machine learning and embedded systems.
An end-to-end data pipeline and analytics dashboard running across AWS and Cloudflare, with every resource declared in Terraform.
Infrastructure as code was the point of the build. A single terraform apply stands up the entire stack from an empty AWS account — three S3 buckets, IAM roles and least-privilege policies, the Glue catalog, an Athena workgroup, the Lambda publisher, the R2 bucket and the Worker deployment. No console clicking, no undocumented setup steps.
A dashboard that turns competitive match history into coaching recommendations grounded in a player's own data.
A Random Forest predicts win probability from prior-match rolling averages, with SHAP explanations surfacing which factors moved each prediction. Evaluation is strictly chronological and features are built from historical data only, so the model is never scored on information it could not have had.
Trained a simulated 7-DOF Franka Emika arm to play mini-golf — grasping, aligning, swinging and striking — as a sequential control problem in a continuous action space.
Two PPO pipelines built in parallel: a Stable-Baselines3 agent tuned for fast, reliable training, and a PyTorch implementation written from scratch for finer control over the policy. Curriculum learning and reward shaping moved the agent through task phases, reaching a 72% success rate across 50 evaluation episodes.
An OCR-driven tool built to shorten insurance onboarding: extracting structured fields from government-issued IDs and populating internal forms with minimal manual entry.
The work covered field-level validation, mapping extracted values to form schemas and secure handling of personal data under privacy requirements, with backend API integration designed alongside it.
An accident-detection system reading sudden shifts in sound level and light intensity to trigger an alarm and dispatch an emergency alert over GSM or Wi-Fi.
A second build automates irrigation from soil-moisture and temperature readings, layering a scheduled baseline over condition-triggered watering and logging sensor data for remote monitoring.
Select an area to see the tools and the work behind it.
Provisioning and operating cloud resources as version-controlled code rather than console configuration. Built and deployed a multi-cloud serverless platform reproducible from an empty account with a single apply, including least-privilege IAM policy authoring.
Designing ingestion and transformation pipelines, modelling data for analysis, and auditing incoming datasets against defined standards — resolving inconsistencies in structure, naming and metadata before they reach downstream work.
Evaluating LLM outputs against structured rubrics, designing assessment frameworks that hold up across distributed review teams, and building supervised and reinforcement learning models with an emphasis on honest evaluation.
Building and maintaining full-stack, containerized applications in a research environment — gathering requirements from non-technical users, writing the documentation that lets them work independently, and shipping through Git-based workflows.
Fragments from the projects — architecture, infrastructure, models and evaluation.
Across AI data, research software, insurtech and university administration.
Verified badges — each links to its Credly credential.
Also — Google Analytics Certification (Google, 2025) · Machine Learning using Python (Simplilearn, 2025) · Data Science 101 (IBM / Cognitive Class, 2025) · Introduction to Cybersecurity (Memorial University, 2024)
If you're building something that needs pipelines, infrastructure or models that hold up in production — I'd like to hear about it.