Production systems, generative AI research, and applied ML — from Snap Lenses reaching millions to autonomous vehicle perception at CMU.
6
Total projects
2
Active
2
In production
2
Published
CATEGORY
STATUS
6 / 6 projects
Snap Inc. · Dec 2025 – Present
Active
Diffusion model fine-tuning for Snap Lenses
LoRA + PEFT adapters on FLUX & QWEN for domain-specific Image2Image tasks. NAS-searched GAN distillation for on-device mobile inference. Piloted face effect Len
↗ 1M+ users
FLUXQWENLoRAPEFTGANNASMobile ML
Snap Inc. · 2025 – Present
Active
Prompt mining framework
Modular config-driven pipeline for image edit, style, video, and txt2image generation. Taxonomy generation from specs with automated prompt harvesting.
↗ Internal tooling
PythonFLUXConfig-drivenTaxonomy
NeuroPixel.AI · 2024 – 2025
Production
FitDiT — flat-lay to model translation
Novel approach converting flat-lay apparel images to model images using FitDiT architecture. Refined apparel masks to improve garment alignment and realism. Upd
↗ Production
PyTorchTransformersCUDADFT Loss
NeuroPixel.AI · 2024 – Present
Production
Real-time segmentation pipeline
Production-grade object segmentation using Meta's SAM optimised to TensorRT format. Achieved 2.69× faster inference for real-time apparel segmentation with scal
↗ 2.69× speedup
SAMTensorRTCUDADockerKubernetesFastAPI
CMU RPAD Lab · 2020
Published
Active perception using light curtains
Collaborative research on perception systems for autonomous vehicles using light curtain sensors. Self-supervised reward functions + Graph CNNs. 80%+ accuracy o
Self-driving robot navigating to GPS waypoints through lanes while avoiding obstacles. EKF-based localisation, motion controls, and electronic architecture. 2nd