I’m a Computer Engineering student at NYU Abu Dhabi, currently studying away at NYU Tandon, with a minor in Applied Mathematics. My work spans efficient LLM inference, autonomous navigation, robot vision, MNAR time-series modeling, quantum sensing, and full-stack scientific software. I build systems end-to-end: from hardware-aware ML evaluation pipelines and visual SLAM navigation stacks to NV-center quantum sensing tools, PK/PD simulation platforms, and reproducible probabilistic modeling code.
Lightweight request-boundary controller for single-GPU LLM inference that routes each request across fixed inference modes including FP16, GPTQ 4-bit, INT8 quantization, speculative decoding, prefix caching, continuous batching, and hybrid configurations. Evaluated on Meta-Llama-3.1-8B-Instruct with vLLM on a single NVIDIA A100 GPU.
MNAR-aware state-space modeling for traffic sensors, treating blackout patterns as an informative observation channel. The work separates blackout-window imputation from post-blackout forecasting and benchmarks against LOCF and MAR LDS baselines.
My work sits at the intersection of AI, robotics, experimental sensing, and scientific software. Recent projects include large-scale migration–hazard modeling, efficient LLM inference, robot vision, autonomous navigation, NV-center quantum sensing, and full-stack research tools. Across these projects, the common thread is building systems that connect theory, data, hardware, and deployment — whether that means a quantum diamond spectromete with experiment-control software, a PK/PD simulation platform for synthetic biology, or navigation stacks that turn sensor data into reliable autonomous behavior.
🏅 iGEM 2025 Gold Medal and Best Integrated Human Practices: Recognized for CDG-focused synthetic biology work and stakeholder-informed project design.
Engineered a full-stack quantum diamond spectrometer setup for NV-center sensing.
Produced emergency-response routing and recharge strategies for Koliwada using field surveys, GPS, and open mapping.
Led 15+ cross-department projects improving water access, Wi-Fi, laundry, sustainability, and student-facing infrastructure using feedback-driven systems.
Workload-aware request-boundary controller for single-GPU LLM inference, routing across FP16, GPTQ 4-bit, INT8, speculative decoding, prefix caching, and hybrid modes. Achieved 2.10× mean latency speedup and 0.48× energy ratio vs FP16 on synthetic workloads while preserving benchmark quality.
End-to-end hardware navigation stack combining 2D LiDAR occupancy-grid mapping, wheel-encoder odometry, AprilTag landmark localization, A* planning, corridor centering, and emergency obstacle stopping.
Three-layer autonomous navigation system for the AI4CE Visual Navigation Challenge using exploration images, live FPV input, offline visual-geometric mapping, RootSIFT-VLAD localization, epipolar target refinement, and direction-aware A* planning. Completed the maze with 0.051 m final translation error.
MNAR-aware linear dynamical system for traffic sensors, combining EKF, RTS smoothing, and EM training to model state-dependent blackouts. Evaluates blackout-window imputation separately from k-step post-blackout forecasting.
PyQt6 interface for configuring NV-center quantum sensing experiments through YAML templates, code-free run launch, automated pulse-sequence setup, and real-time ODMR fitting.
Web app for PK/PD fitting and simulation with 1–3 compartment models, KPI extraction, dosing-regimen builder, VPCs, and automated PDF export.
PyTorch CycleGAN pipeline for unpaired cultural style transfer between Bandhani textile motifs and modern-art patterns, with preprocessing, adversarial/cycle-consistency training, checkpoint loading, and a Gradio demo.
MySQL-backed airline reservation system with a Flask web app, supporting flight search, ticket purchase, trip history, and staff dashboards on a normalized relational schema.
Achieved 700×–2000× lossless compression of traffic matrices using feature-based sampling and matrix completion. Implemented full encoding, decoding, and BCD optimization in NumPy/SciPy.
PyTorch DCGAN trained to generate Fashion-MNIST clothing from latent vectors, with full adversarial training pipeline, loss tracking, and checkpointed models.
Simulation tool for ODMR spectra and NV-center sensing experiments, built to accelerate quantum-sensing validation and workflow development.
Voice-navigated 3D campus tour of NYU Abu Dhabi built in Unity with teleportation, voice transcription, and interactive scene switching.
Designed and implemented a 14-bit signed microprocessor in VHDL with a 16-instruction ISA, ALU, branching, and seven-segment FPGA display.