Arya Lyngdoh Lakshmanan
- Location: Brooklyn, NY
- Links: Email, LinkedIn, Personal Website, GitHub
Education
New York University / May 2026
- M.S. Scientific Computing
- Focus: High-performance modeling/forecasting
- Courses: Numerical Methods, Machine Learning, Deep Learning, GPU Architecture & Programming
Rutgers University / May 2023
- B.S. Astrophysics & Financial Economics
- Awards: Magna Cum Laude, Highest Honors, Aryabhata Endowed Award
- Thesis: Statistical modeling of dark matter and supernovae
Skills
- Languages: Python, C++, CUDA, Rust, JavaScript, GLSL
- ML: PyTorch, Transformers, Model Evaluation, Burn (PyTorch equivalent)
- Systems: Docker, Linux, Bash, Git
- Data: SQL, Pandas, NumPy, R
Work Experience
AI and LLM Consultant / WordsworthTech Inc
Dec 2023 – Apr 2024, May 2025 – Aug 2025
- Architected a Node.js API to orchestrate low-latency interactions between the StreamAlive platform and LLMs (Gemini/OpenAI), optimizing for high-concurrency live session environments
- Developed custom parsing logic and validation layers to handle stochastic LLM outputs, ensuring strict type-safety and reliability for downstream application logic
- Established quantitative metrics to monitor token usage and response quality, implementing caching and prompt-tuning strategies that reduced operational costs while maintaining model performance
Undergraduate Research Assistant / Rutgers University
Aug 2021 – May 2023
- Researched the effect of frequent supernovae on the dark matter within dwarf galaxies under Dr. Kristen McQuinn (NASA Roman Space Telescope Mission Head)
- Developed regression models in Python to estimate dark matter distributions in dwarf galaxies
- Engineered automated testing routines for galaxy simulations, reducing computation time and minimizing error margins in large-scale dataset analysis
Projects
Multi-GPU Performance Analytical Model | CUDA, C++, Python
Sep 2025 – Dec 2025
- Quantified compute vs communication tradeoffs to identify scaling limits under PCIe constraints, informing when multi-GPU parallelism degraded performance
- Implemented micro-benchmarks in CUDA to calibrate hardware parameters (PCIe throughput, FLOPs) on arbitrary Nvidia architectures
- Validated the model against bottlenecked algorithms (N-Body, Conjugate Gradient), successfully identifying hardware saturation points and scaling limits
Rust/CUDA Simulation Engine | Rust, wgpu, CUDA
Sep 2025 – Present
- Re-engineered a solar system simulation engine in Rust, migrating from my prior JavaScript WebGL/GLSL implementation to a Rust WebGPU API
- Accelerated procedural texture generation by implementing a specialized version of Perlin Noise in CUDA
Rust-Based Deep Learning Audio Classifier | Rust, Burn
Aug 2024 – Present
- Engineered an asynchronous, multi-threaded deep learning pipeline in Rust using the Burn framework to classify audio features with Transformer-based attention mechanisms
- Implemented a classification batcher with dynamic padding masks to efficiently process variable-length sequential data, optimizing memory usage during inference
Societies and Extracurriculars
President and Treasurer / Rutgers Astronomical Society
Sep 2021 – May 2023
- Managed budget and led weekly public seminars for 100+ attendees; served as department liaison for university events.
- Created and maintained an LGBTQ+ safe space in collaboration with Rutgers minority advocacy groups.