Projects

Performing LWFA is highly sensitive to plasma parameters, and each laser system requires a tailored gas plume with well-defined spatial extent and density

• Synthesized academic literature to parse LWFA scaling laws and inform Ansys Fluent CFD + WarpX PIC simulation parameters
• Developed bash scripts to leverage Linux-based national Fir supercomputing cluster for 100x faster simulations
• Scaled Python data analysis suite for optimized review and comparison of >300GB datasets
• Designed and performed optical experiments, working with sensitive optics and gas lines, for interferometric validation of gas nozzle
• Presented weekly to research group and visiting members ranging from non-technical to postdoctoral levels

• Successful interferometric characterization of gas nozzle in moderate agreement with simulation results, achieving 75% expected density

Battery voltages were unknown during autonomous operation and battery enclosure leak checks relied on human judgement, creating Li-ion safety risk

• Designed a low-power, multi-board battery safety system in Altium Designer with a magnet-activated display and STM32-based pressure, temperature, and voltage DAQ, cutting under-voltage incidents by 90%
• Engineered power delivery and IC operating modes, validating <300 µA passive load against datasheet specifications via strict component-level testing
• Optimized C firmware in STM32CubeIDE with DMA, interrupts, and sleep mode, averaging 315 µA current draw
• Headed full-stack PCBA design: low-power part selection and sourcing, schematic capture, layout, manual assembly, bring-up, C++ firmware, and system integration testing
• Authored custom footprints + symbols respecting manufacturer capabilities for: STM32L072CZ, QFN buck converter, various passives
• Routed controlled-impedance USB 2.0 to manufacturer stackup using differential-pair and via-stitching tooling, verified by successful USB-C DFU flashing

• Optimized space-constrained battery pod PCB system for 315 µA average current draw
• Implemented firmware pressure-testing sequence to catch 100% of user errors in high-traffic battery system
• Improved ease of data reporting reduces under-voltage incidents by 90%

Need to accurately emulate high-energy electron radiation found in space to calibrate a satellite-mounted particle detector for climate research

• Automated FEMM electrostatic simulation of various cathode geometries using Lua and Python co-integration
• Vectorized Python relativistic Runge-Kutta 4 electron trajectory physics engine to simulate 1000+ electrons through 20000 steps in <10s on consumer laptop
• Constructed high-voltage Cockcroft-Walton multiplier to achieve >50 kV, >10 MV/m accelerating gradient
• Synthesized vacuum breakdown literature to analyze arcing models, eliminating 90% of 'noisy' electrical discharge to permit reliable data acquisition
• Used SolidWorks and shop tools to design and fabricate a mount for vacuum chamber diode validation

• Successful 50 keV electron generation; outstanding poster award among 107 submissions

• Managed feature branches on GitHub so five contributors could develop in parallel without blocking hardware time
• Integrated an MPU-6050 IMU, ultrasonic rangefinder, 8-bit display mode select, servos, and motors on a C + PlatformIO embedded stack
• Designed mission planner system hierarchy with PD controller module accepting position and heading targets

• 1st prize; only team to achieve 100% completion of all competition tasks

Commercial CV research drones cost thousands and lack open-source firmware to enable industry-standard ROS 2 development, making them unviable for a research environment

• Engineered ROS 2 + OpenCV vision pipeline for 1080p60fps video with <160ms capture-to-processing latency
• Leveraged open-source foundations, forking ArduPilot C++ firmware to enable micro-XRCE DDS, allowing ROS 2 communication on telemetry network
• Achieved 4x cost reduction over commercial model via careful system design and SolidWorks modelling of modular structural components
• Configured optical flow and rangefinder for stable hover, holding <10cm drift over 5+ minutes
• Wrote technical design documentation and continued support of undergraduates on the project
• Honed technical hardware assembly (crimping, soldering) and debugging

• 1080p60fps video with <160ms capture-to-processing latency at 4x cost reduction over existing model

A Clearpath Jackal running ROS1 Noetic on legacy hardware had no autonomy capabilities, requiring migration to ROS 2 and custom algorithm development

• Engineered system hierarchy to use dynamic global Dijkstra search over an occupancy grid from Steve Macenski’s slam_toolbox for 100% goal convergence
• Added Python data-processing stage, using cubic-spline smoothing with adaptive downsampling to reduce waypoint density, limiting jitter
• Designed a local controller with an MPC loop to respect Jackal acceleration limits, smoothing velocity profile
• Rebuilt firmware and ported Nvidia AGX Xavier-based platform to ROS 2 Humble, compiled Velodyne VLP-16 and FOG IMU drivers from source, and containerized with Docker
• Validated SW in Gazebo + rviz2 simulation before hardware runs to catch faults

• Sub-10 cm cross-track error and 100% goal convergence across hardware trials
• Runs as the lab's default autonomy baseline for follow-on projects

• Engineered OpenCV + Python custom tracker, fusing Circle Hough Transform with motion-based prediction model for tracking resilience across 5+ consecutive failed segmentations
• Tuned PID to control Tello SDK velocity commands; designed communication model to fail safely, resulting in drone auto-landing if connection drops

• Designed hardware-redundant magnetic killswitch PCBA ensuring diver safety with 100% AUV kill rate on 500+ triggers
• Completed magnetic field strength threshold calculations to ensure 100% consistency; meets all RoboSub competition safety requirements
• Authored custom component footprints + symbols in Altium Designer, standardized against workspace requirements and manufacturer clearance constraints

• Modelled and fabricated Archimedes' screw geometry with belt-driven generator, powering an LED
• Designed model in SolidWorks and rendered in Blender; fabricated from 3D printed, laser-cut and heat moulded components

• Achieved near-perfect score across design documentation and output power performance

• Coded calibration sequence to perform dynamic statistical analysis of startup sensor readings to improve low-contrast operation
• Tuned C-based PID controller to reduce robot jitter, resulting in <1cm overshoot
• Modelled and 3D-printed custom CAD frame in SolidWorks for sensor mounting and wiring harness

Build a budget-constrained robotic arm to sort plastic balls

• Led a team of 5 to iteratively prototype 4DOF claw machine-inspired manipulator
• Repurposed recycled materials (scrap wood, PVC, thread spools) to spend <$5 on components excluding motors

• Achieved the Gold Standard (highest competition tier), successfully completing all competition tasks

• Optimized minimax search algorithm with alpha-beta pruning to reduce game tree exploration time by >60%
• Used Python to implement recursive game loop and lightweight Flask REST API

• The agent is very challenging to beat at a recursive depth limit of n=8