— Portfolio —

Projects

Fifteen projects spanning renewable energy systems, mechanical engineering research, and AI training infrastructure — from feasibility studies for remote communities to rubric design for frontier AI models.

Engineering (8) AI (7)

Engineering

8 Projects

2023 – Ongoing

Geothermal Hybrid Model for Diesel Replacement

Engineering Current

Active research designing a geothermal + renewable hybrid energy system as a sustainable alternative to diesel generation in remote Newfoundland and Labrador communities. The work combines low-enthalpy geothermal sources with solar PV, wind, and battery storage to deliver community-scale, weather-resilient power.

Focus areas include techno-economic feasibility, lifecycle cost analysis, and configuration optimization under realistic load profiles. The goal is to reduce dependency on costly diesel fuel transport, lower lifecycle emissions, and improve energy resilience for Indigenous and rural NL communities currently reliant on diesel generators.

Tools: HOMER Pro MATLAB PLEXOS

Jun 2023 – Ongoing · Memorial University

Pre-Feasibility Hydrogen Production, Storage & Distribution for NL

Engineering Research

Comprehensive pre-feasibility study assessing hydrogen production, storage, and distribution potential across Newfoundland and Labrador. Evaluates scalable, sustainable pathways aligned with provincial energy goals and Canada's wider hydrogen export ambitions.

Scope covers source-to-application supply chain modelling, electrolysis technology selection, storage method comparison (compressed gas, liquid hydrogen, ammonia), and economic viability under multiple cost and demand scenarios. Findings inform investment and policy direction for the province's hydrogen economy.

Tools: HOMER Pro PLEXOS Excel

Memorial University

Hydrogen Production System Modelling

Engineering Research

Computational research and PLEXOS-based modelling of hydrogen production systems, with applied analysis on the Canada-to-Germany supply chain pathway. Examines production technology options, transport modes, and conversion considerations along the supply chain.

Model captures end-to-end production economics, energy carrier choices (liquid H₂ vs. ammonia vs. methanol), and capacity-to-demand matching. Supports strategic decisions for international green hydrogen export from Canadian renewables to European markets.

Tools: PLEXOS Excel MATLAB

Jun 2023 – Ongoing · Memorial University

Microgrid Projects for Local Communities in Newfoundland and Labrador

Engineering Research

Leading the development and optimization of microgrid systems combining wind, solar, and hydropower to provide sustainable energy solutions for local NL communities — including hybrid renewable energy analysis for Nain, Hopedale, and Natuashish.

Each community-specific configuration involves load profile analysis, system sizing, dispatch optimization, and 20+ year lifecycle planning. The goal is reliable, locally appropriate renewable systems that displace high-cost, high-emission diesel generation while accounting for unique geographic and climate constraints of each location.

Tools: HOMER Pro PLEXOS

Oct 2022 – May 2023 · Master's Thesis

Numerical Study of Turbulent Flow Around Submerged Tidal Turbine Foundations

Engineering Master's

Computational fluid dynamics analysis of hydrodynamic forces on submerged tidal turbine foundations. Used ANSYS Fluent to model turbulent flow patterns, vortex shedding, and structural loading under varied current conditions.

Outcomes informed optimized foundation designs that enhance turbine efficiency, reduce sediment scour, and lower fatigue risks. The work contributes to safer, more efficient tidal energy deployment in coastal environments.

Tools: ANSYS Fluent CFD MATLAB

Jul – Oct 2022 · Master's

Design of a Solar Grid-Tied System for Residential Use in NL

Engineering Master's

Developed and implemented a solar grid-tied PV system for residential application in Newfoundland and Labrador. Used PVsyst for system sizing, performance simulation, and shading analysis tailored to NL's high-latitude solar profile.

The final design delivers measurable reduction in residential carbon footprint and significant long-term cost savings under NL's electricity pricing structure, while accounting for snow load, panel tilt optimization, and net-metering integration.

Tools: PVsyst AutoCAD

Jun – Sep 2022 · Master's

Optimization of Hybrid Microgrid System in West Kelowna

Engineering Master's

Techno-economic optimization of a hybrid microgrid for a small-scale community in West Kelowna, British Columbia. Compared multiple configurations of solar PV, wind, battery storage, and grid-connected backup.

Achieved a cost-efficient, reliable, and eco-friendly 20-year energy supply with sensitivity analysis on equipment cost, fuel price, and load growth. Final design balances LCOE, NPC, and emissions targets for sustainable community-scale energy.

Tools: HOMER Pro Excel

Jun 2019 – May 2020 · Lassol Private Limited, Coimbatore

Design of an Industrial Hopper for the Restaurant Industry

Engineering

Developed a high-capacity industrial food hopper for the restaurant industry in Coimbatore, India. Designed in SolidWorks to meet stringent food safety standards, reduce food waste during processing, and enhance kitchen productivity.

Work included material selection, finite element stress analysis, and manufacturing tolerance specification to ensure the unit performed reliably under continuous commercial use.

Tools: SolidWorks FEA AutoCAD

AI

7 Projects

Python · JSON Rubric Schemas

AI Training Rubrics for Mechanical Engineering

Designed reference plans and JSON-based grading rubrics that train and evaluate AI coding agents on mechanical engineering software tasks. Wrote planning statements that mirror real senior-engineer requests, authored ideal reference plans, and built two paired rubrics per task: a planning rubric (grading plan quality) and an execution rubric (grading the final code).

Every criterion was written to be objective, self-contained, and grounded in repository conventions — avoiding vague language and stacked checks. Rubrics were calibrated for a sub-60% frontier-model pass rate, then cross-validated for alignment between the planning statement, golden plan, golden patch, and both rubrics.

Tools: Python JSON Cursor Git/GitHub

Python · Agentic Tools

Codebase & Repo Reasoning Tasks

Built Python-based AI training tasks focused on codebase reasoning and exploration. Tasks teach AI agents to navigate, understand, and extract information from real software repositories — replicating the kind of investigation an onboarding engineer would do on day one.

Each task includes deterministic verifiers, structured expected outcomes, and rubrics for grading agent behavior across exploration depth and reasoning quality.

Tools: Python pytest Agentic coding tools

Python · Multi-step Benchmark

Multi-Step Scientific Coding Benchmark

Built multi-step Python coding tasks for an open-source benchmark used to evaluate frontier AI models on real research-paper-derived problems. Each task is derived from a recent research paper introducing a novel computational method, decomposed into 6–10 unit-testable subproblems, and implemented in clean, self-contained Python.

Work involved sourcing papers with novel algorithms, decomposing methods into incremental Python steps, writing strict unit tests with calibrated floating-point tolerances, and ensuring reproducibility. Tasks were calibrated to be genuinely challenging — most frontier models solve under 10%.

Tools: Python pytest Docker GitHub PR workflow

Python · Extended Contributions

Scientific Coding Benchmark — Extended Work

Continued contributions to the scientific coding benchmark with additional Python-based multi-step tasks. Focused on numerical methods, scientific algorithms, and reproducible implementations.

Strict unit tests with calibrated floating-point tolerances and fully Dockerized environments ensure reproducibility across machines and reviewers.

Tools: Python pytest Docker

Review · Rubric Design

Rubric Writing for AI Coding Tasks

Analyzed existing Python-based AI coding tasks written by other contributors and authored planning and execution rubrics. The role focused exclusively on rubric design — reading task specifications and solutions, then writing objective grading criteria across functional, robustness, and style dimensions.

Rubrics were calibrated to challenge frontier models and validated for alignment with task requirements and reference solutions.

Tools: Python JSON Code review

Writer · Whole-Task Authoring

Codebase Exploration RL Environment

Designed and built RL training tasks for an environment that teaches AI agents to explore and reason about unfamiliar open-source software repositories — the kind of high-level engineering questions an onboarding engineer would ask their tech lead.

Each task posed a non-trivial natural-language question about a real open-source repo, paired with a deterministic pytest verifier over the agent's exploration trajectory and an LLM-as-Judge rubric for grading the final answer. Produced golden trajectories by solving each question manually using agentic coding tools.

Tools: Python pytest OpenCode LLM-as-Judge

Writer · Reviewer · Super Reviewer

Behavioral Failure Mode Benchmark for AI Coding Agents

Contributed to a 500-task benchmark designed to systematically measure behavioral failure modes in AI coding agents — cases where a model has the capability to solve a task but doesn't reliably do so. Built and reviewed Dockerized tasks targeting specific failure modes: poor time management, analysis paralysis, wrong tool selection, insufficient verification, and perseveration.

Each task included a deterministic golden solution and verification tests. Served as writer, reviewer, and super reviewer — calibrating difficulty, auditing across 11 quality dimensions, and resolving issues with direct PRs.

Tools: Docker Python pytest Code review