Uncover the performance of AI models with AI + 150,000 global evaluators.

Community based AI model red-teaming. YC W22

Helped build self-driving Pikachu

We made a new approach to push the technical and user experience boundaries around computer vision, deep learning, augmented reality, and robotics. Led a team of engineers to build a unified mapping platform for augmented reality and robotics.

Led the design and implementation of highly accurate bundle adjustment pipeline for aerial imaging systems for Google 3D Maps. Full software development cycle from planning to architecture design, implementation, testing, launching, and maintaining.Proposed high resolution 3D reconstruction by Google internal satellite imagery. Formed and led cross team efforts, now a part of the imagery group business plan.Increased pipeline throughput by 20x.Received 6 peer and 2 spot bonuses for project excellence and leadership.

- Investigated multiview scene reconstruction framework, proposed scene structure modeling for 3D viewing.- Developed a new multiview stereo matching technique - Developed multi-label optimization of Markov random fields model and applied to a novel image inpainting technique.

Devised a multidimensional incremental parsing algorithm for multidimensional discrete sources as a generalization of the Lempel-Ziv incremental parsing. Developed a feature extraction technique based on a universal source coding and the application to semantically higher-level representation of multidimensional source, and applied to content-based image retrieval (CBIR)

Developed a perceptually-driven low bitrate quantization technique for a HiFi audio coding. Published 1 conference paper.

Investigated human visual system response to scalable image/video compression. With this result, we published 1 conference and 1 journal papers.

Developed optimized software implementations of G.729 voice coder and H.263 video coder for commercialized video conferenceing system.