arissa zheng
about experience gallery projects recordings
Projection Spaces Coming Soon

Walk-through rooms projected from a single photo

Building a projection-based installation that turns single photos of rooms into spaces visitors can walk through and interact with, using monocular depth estimation (Depth Anything V2), object segmentation (SAM 2), and real-time head and hand tracking to render head-tracked parallax and gesture-driven object manipulation on a projected wall.

Moodboard.map Coming Soon

AI design generation with your taste from the physical world

Building an AI design tool that turns everyday inspiration from the physical world into a design taste engine: designers snap photos of what catches their eye, and the app extracts colors, patterns, and textures to learn their taste, then guides generative AI to create assets that match it instead of generic output.

Autostitching and Photo Mosaics Nov. 2026Link

Automatic panorama stitching, built from scratch

Built an automatic panorama-stitching pipeline in Python/NumPy, implementing Harris corner detection, adaptive non-maximal suppression, feature descriptors with ratio-test matching, and 4-point RANSAC homography estimation from scratch to align and blend photos into seamless mosaics.

Flow Matching Image Generator Oct. 2026Link

Flow-matching image generation from scratch in PyTorch

Built and trained time- and class-conditioned UNet flow-matching models from scratch in PyTorch on MNIST, and implemented Euler ODE sampling with classifier-free guidance on a pretrained 512×512 text-to-image model for text-guided editing, visual anagrams and hybrid images.

Niche Oct. 2026Link · Github

Agentic shopping with your taste profile

Agentic shopping with your taste profile. CalHacks Best Use of Capital One Nessie API.

Fit Game Oct. 2026Link · Github

Dress to Impress, with your own wardrobe

Dress to Impress, with your own wardrobe transformed into playable assets. CalHacks Women Hack Best Overall Prize.

Image Blender Sept. 2026Link

Seamless image blending in the frequency domain

Implemented frequency-domain image-processing algorithms (Gaussian/Laplacian stacks, unsharp masking, hybrid imaging) from scratch in NumPy to seamlessly blend images with irregular masks, validating custom convolution against SciPy.

Image Colorizer Sept. 2026Link

Colorizing historical glass-plate photographs

Designed and implemented a coarse-to-fine image registration algorithm (Gaussian pyramids, NCC/L2 scoring) to automatically colorize historical RGB glass-plate photographs, vectorizing the full pipeline in NumPy for sub-minute runtime on large TIFF inputs.

Secure File Sharing System Mar. 2026

End-to-end encrypted file sharing in Go

Developed a secure file-sharing system in Go with the ability to detect 100% of attacks on a vulnerable database using Symmetric/Asymmetric/Hybrid encryption, MACs, Digital Signatures, etc.

Axis Dec. 2025Link · Github

Make any screen gesture-controllable

Built OpenCV-based software to make any screen gesture-controllable with swipes, taps, and zooms.

GeoMeter Oct. 2025Link · Github

The carbon footprint of your AI prompts, in real time

Built and designed a Chrome extension that estimates energy, water, and carbon footprint of your AI prompts in real time and suggests greener alternatives.

freeventures.org Aug. 2025 – PresentLink · Github

Website for Berkeley’s zero-equity startup accelerator

Maintaining website for Free Ventures, Berkeley’s zero-equity startup accelerator.