Danyal Khorami

Mesquite MoCap is a real-time, full-body motion capture system built with 15 wireless ESP32-C3 IMU nodes. Designed for accessibility and open-source transparency, it achieves 32 FPS capture rate with 15ms latency and 99.7% packet delivery. The system exports to BVH format and has been benchmarked against OptiTrack (±2.5° joint error).

Technical Highlights: Arduino/ESP32 firmware with time-synchronized networking, real-time sensor fusion algorithms, WebXR integration for live preview, optimized serial communication for embedded constraints.

Developing a scalable multi-view 3D capture system using 4 synchronized Android phones (Samsung S23 class) for accessible human motion capture and XR asset creation. The system performs real-time multi-view calibration, temporal synchronization, and outputs ready-for-VR meshes.

Key Focus: Low-cost accessibility, precise multi-camera synchronization, robust pose estimation pipelines, rapid mesh reconstruction for real-time XR integration.

"To Wilt" is a research-driven 4-space installation exploring how Large Language Models express, simulate, and question emotion through extended dialogue and sensor-fusion environments. The work challenges anthropomorphic AI assumptions while celebrating human temporal perception.

Spaces: (1) 8-screen video wall with 3D rose-growth animation + fine-tuned LLM philosophical dialogue, (2) Real-time LLM conversation printed live on networked dot-matrix printers, (3) 22k-image time-lapse decay dataset with LLM analysis, (4) Sensor-instrumented rose tank reflecting emotional decay through environmental metrics.

Happenstance is an end-to-end pipeline for converting single human photographs into rigged, animation-ready 3D characters for VR/XR environments. Uses Facebook Research's PiFUHD for high-resolution implicit function-based 3D reconstruction, followed by automated rigging in Mixamo.

Current Development: Improving mesh quality post-processing, automating Mixamo integration, reducing end-to-end latency for real-time applications, exploring diffusion-based clothing and accessory generation.

Opuntia is an off-grid environmental monitoring station powered by solar panels and Li-ion battery stack, built with ESP32-C3 microcontroller. The system integrates multi-sensor telemetry (temperature, humidity, pressure, soil moisture) with MongoDB cloud logging and local measurement archiving for microclimate analysis.

Technical Achievement: Power budget optimization, duty cycling for extended uptime, timestamp integrity during outages, automated backfill protocols, real-time website broadcasting of environmental data streams.

Completed rigorous deep learning coursework (EEE 598: Deep Learning Foundations & Applications) with emphasis on large-scale model training. Projects include Vision Transformers on PV-fault classification, custom ResNet architectures with novel activation functions, GANs and diffusion models with latent interpolation studies, and perceptual loss design using VGG-16 features.

Technical Depth: Advanced PyTorch, CUDA optimization, multi-GPU training strategies, Grad-CAM visualization, ablation studies, model architecture design from first principles, benchmarking on NVIDIA A100 vs. Intel Habana Gaudi through SLURM HPC workflows.