- What I ledDerived the thermodynamic equations for the HTF7000 turbofan radiator redesign and validated the model with teammates.
- ToolsBuilt optimization scripts in Python to tune heat-transfer performance across dozens of design permutations overnight.
- OutcomePlaced 4th overall and earned a $1k prize, plus an encore presentation to Honeywell engineers.
All Projects
YearProjectMade atBuilt withLink
- Lab ProtocolsTrained in multiple lab protocols, including radiation safety and sample handling. I learned analysis techniques such as Rutherford Backscattering Spectrometry (RBS) and Particle-Induced X-ray Emission (PIXE).
- Hardware MaintenanceUnder supervision, I learned to open and perform maintenance tasks on high-vacuum systems. I also assisted in troubleshooting the ion beam accelerator, helping disassemble components to understand its mechanics.
- Data AnalysisMastered the primary software tools for RBS data analysis, including xRUMP and SIMNRA. I later used these skills to analyze data from NASA's Genesis mission.
- GoalAdapted and optimized a RealESRGAN (super-resolution) model for high-speed, low-resource inference on edge devices (Android).
- Tech StackUsed the Qualcomm Neural SDK and a hardware-aware back-end to accelerate the model.
- PerformanceAchieved a sub-100ms processing time to upscale images by 10x, a significant performance leap for on-device super-resolution.
- RoleOwned the LLM pipeline end-to-end: curated datasets, fine-tuned TinyLlama, and deployed the inference stack for the mobile app.
- CollaborationCoordinated asynchronously across time zones to align AI sprints with UI milestones and hand-offs.
- ImpactWon the hackathon and produced a working sustainability coach that ships daily nudges and free chat responses.
- AIFine-tuned Google’s BERT model on Kaggle datasets to flag scam messages with production-ready accuracy.
- ProductDesigned and built the end-to-end UI/UX for both the web app and the public landing page.
- ProcessKept a distributed team in sync across 12+ hour time differences with async updates and clear specs.
- My RolePitched the initial concept and led the back-end development, integrating multiple AI and OCR APIs.
- Tech StackUsed React-Native for the front-end, while the back-end analyzed rubrics with GPT-4 and extracted text using Tesseract and Nougat OCR. I tried using Flask in back-end as well, but eventually settled on JS.
- Key LearningA high-pressure, 2-day project that solidified my back-end and API integration skills. It was a challenge I'm glad I took on.
- GoalBuilt a Convolutional Neural Network (CNN) to classify MNIST handwritten digits, achieving 99.2% accuracy.
- ProcessStarted by coding the network's layers from scratch, then progressed to using standard libraries to experiment with different optimizers (SGD, Adam) and loss functions (CrossEntropyLoss).
- OutcomeThis project was a deep dive into the fundamentals of CNNs. I wrote a comprehensive, step-by-step guide explaining the math and code.