Projects

Note: This page is still under construction and is not fully updated!

Hearti (Devpost, Slides)
Hearti is a tool for doctors and patients to make diagnosing and treating congenital heart disease easier to do and explain. We leverage the HVSMR2.0 dataset to inform new treatments with past interventions on patients with similar heart problems. First, we convert a set of CMR images to a segmented 3D model. This model is converted to embeddings and stored in Vespa. New sets of CMR images are converted to 3D, then embeddings, and queried in Vespa to find the most similar heart model. The associated metadata of the best match is passed to Perplexity to generate an evidence-based diagnosis and treatment plan, which is then used by an ElevenLabs conversational agent to enable patient Q&A. You can use this tool on our website, and the 3D model can also be interacted with in an Oculus headset.

We built Hearti in one weekend at Stanford’s TreeHacks 2025. Hearti won the Most Impactful Grand Prize out of 270 projects and 1,000 hackers!

Face Morphing (Website)
I produce “morph” animations of faces into different faces, compute the mean of a population of faces, and change the ethnicity of a friend’s face using only classical computer vision techniques.

Autostitching Photo Mosaics (Website)
Casual imagery of a scene is not necessarily captured in a way conducive for panorama creation. I use photographs of a scene to create image mosaics by registering, projective warping, resampling, and compositing them. I compute homographies to warp images, then use a Laplacian stack to blend images into the final mosaic. I then automate correspondence-finding by implementing Harris interest point detection, followed by adaptive non-maximal suppression to limit the number of interest points while maintaining a good spatial distribution over the entire image, scoring and thresholding to extract good matches, then using random sample consensus to find an ideal homography.

Diffusion Models (Website)
Using Stability AI’s DeepFloyd IF two-stage diffusion model, I implement sampling loops for classical, one-step, and iterative denoising; obtain high quality samples with classifier-free guidance; perform image-to-image translation with the SDEdit algorithm; and create visual anagrams and hybrid images. Then, I train my own diffusion model (using the UNet architecture) to generate digits from the MNIST dataset. I start by building a one-step denoiser, then add time and class-conditioning for iterative denoising.

Lightfield Camera and High Dynamic Range Imaging (Website)
To recreate the effects of the Lytro lightfield camera, I use rectified images provided in the Stanford Light Field Archive and apply alignment and averaging methods to implement depth refocusing and aperture adjustment, allowing for post-capture changes in the camera setup. In order to fix image over/under-exposure, I implement tools to automatically combine multiple exposures into a single high dynamic range radiance map, and then convert this radiance map to an image suitable for display through tone mapping.

Colorizing the Prokudin-Gorskii Photo Collection (Website)
Using three black-and-white negatives, I use image processing techniques to align these plates and produce a color image with minimal artifacts.

Fun With Filters and Frequencies (Website)
I visualize image edge detection and convolution, implement image sharpening, create hybrid images (static images that change while varying viewing distance), and perform multi-resolution blending with Gaussian and Laplacian stacks.

Papers & Publications

From Simulation to Reality with Random Noise
Steven Luo, Rory Lipkis, Ignacio G. Lopez-Francos, Pavlo G. Vlastos and Adrian Agogino
2025 AIAA SciTech Forum: Conference Paper (Journal, PDF) and Technical Presentation (Slides)

Reimagining Open Source and Openness in AI: Co-creating Responsible Technological Futures Genevieve Smith, Steven Luo, Hiral Patel, … Workshop Report PDF

Summit on Responsible Computing, AI, and Society (October 2025, Atlanta, CA) Poster: Inclusive, Interdisciplinary, and Scalable Models for Building University-Centered Responsible AI Communities