Infra TL for the next-generation visual & audio similarly search (AViSS). The 2-year project upgraded the fingerprinting technology from handcrafted to ML-based, and improved recall by +27% (@90% precision). As infra TL, I led 6 engineers and achieved the following:* Param tuning: worked with ML engineers to tune fingerprinting and indexing parameters. Running experiments and to make informed tradeoffs between precision, recall, compute and storage.* Inference infra rearchitect: decoupled decoding and inference. Moved the former to a shared video decoding platform, and the latter to GPU. This yielded 70% efficiency wins.* Retrieval infra rearchitect: moved from sharding by documents (search fans out to all shards) to sharding by features (kNN search nearest centroids). This yielded 50% efficiency wins.* Enhance verification infra: added ability to extend match segments on both ends. Leveraged memcache for scaling.* Fully launched the new fingerprinting technology, for both online (handling new uploads & livestreams), and offline (backfill for the entire video catalog)

As a senior IC in Video Copyright Infra, I launched two critical projects to support Europe Copyright Directives:* Scale up retrieval infra: designed and implemented SnapshotService to offload index building, and further optimized index loading with P2P caching. This accelerated index warm-up time by 10x, and reduced DB load by 100x.* Leverage Elastic Compute: took advantage of traffic shaping through queuing and dynamic back pressure. This reduced the need for dedicated resources by 80%.

As an IC in Search Result Page Blending, I built infrastructure for dynamically placing different types of search results, including people, groups, photos, posts, and etc.

I was a Research Assistant, working with Prof. Eleazar Eskin. My primary research interest is Bioinformatics, more precisely GWAS (genome-wide association studies).

Work in search team, more precisely Web Answer.

I was selected as a TA for Design and Analysis of Algorithms (seminar) while most of other TAs were graduates. And once again, I worked for professor Yingfei Xiong.

I was selected as a TA for Introduction to Computer System (seminar), and the corresponding professors for my class are the professor Yingfei Xiong and the dean Mei Hong.