Experience

• Personalize the YouTube.com home page/screen to your interests by surfacing relevant content.
• Machine learning to facilitate serendipitous discovery of engaging content.
• Improved user engagement and retention on YouTube through better personalized video recommendations.
• Helped make YouTube a destination site by tailoring the home page content and ranking to the individual user by engineering infrastructure for large-scale machine learning recommendation systems.
• Member of Emmy award-winning team!

• The Climate Corporation's mission is to help all the world's people and businesses manage and adapt to climate change. We aim to protect the global agriculture industry from the financial impact of adverse weather with automated, full-season weather insurance.
• Designed and implemented a reliable, latency-critical, RESTful web service running on Amazon Web Services (EC2 and S3) for large-scale analysis of historical weather data and weather forecasting on demand. Use Java, Clojure, and Python to create RESTful back-end web service supporting weather and agriculture tools on the front end.

• Improved user happiness and advertiser return-on-investment and reach by increasing the coverage and relevance of ads on search queries on Google.com and partner web properties.
• Designed and implemented algorithms to improve the targeting of ads, by using semantic knowledge and historical data.
• Implemented algorithms in worldwide live servers with strict latency, CPU, and memory constraints.
• Design, run, and analyze experiments affecting billions of search queries per day and billions of dollars of revenue per year.
• Analyzed performance of ads targeting by implementing statistical algorithms such as hypothesis testing, Bayesian inference, linear and logistic regression.
• Monitor and help maintain production servers worldwide.
• Mentored junior engineers and interns.
• Collaborated with fantastic engineers, research scientists, and product managers in four time zones and three countries and multiple product areas.

• Conducted research in algorithms for combinatorial problems in computational geometry and topology as a scholar in the Center for the Mathematics of Information, Information Science & Technology.
• Collaborated with Caltech researchers in Computer Science on problems in computational geometry and topology, such as those motivated by graphics applications.
• Investigated algorithms for clustering and modeling imprecise data with Caltech colleagues.
• Researched algorithmic problems in topology such as computing optimum homotopy between curves on surfaces and homeomorphism between surfaces, and curve and surface approximation. 

• Advised OpenX on algorithms for design and implementation of a real-time spot auction market for online display advertising as one of two computer scientists on a team of eight consulting for OpenX Inc., a Pasadena-based startup in online advertising.
• Researched algorithms for price support in multiple online Vickrey auctions and for inventory matching to maximize return-on-investment. 

• Researched and implemented software for fast, real-time surface reconstruction from point clouds for dentistry application for Caltech researchers and Arges Imaging. (http://www.caltech.edu/content/3-d-dentistry)

• Devised external-memory cache-efficient and cache-oblivious algorithms for fundamental problems in computational geometry and in optimization. Results are applicable in Geographic Information Systems, where map data is too massive to fit entirely in main memory; Algorithms are practical and implementable as well as theoretically efficient.
• Designed a cache-oblivious IO-efficient version of the algorithm of Frederickson and Johnson for selecting in sorted X+Y matrices. The matrix selection algorithm is a key building block of many other optimization algorithms, sometimes serving as an alternative to expensive parametric search.
• Proved bounds on the combinatorial complexity of Voronoi diagrams on realistic terrains—studied problems in combinatorial geometry. 

• Developed algorithms for spacetime meshing to support novel spacetime-discontinuous Galerkin finite element methods. Our algorithms were the first adaptive geometric algorithms to construct efficiently solvable spacetime meshes. The meshes we construct are suitable for efficient simulation of time-dependent wave-like physical phenomena using spacetime discontinuous Galerkin finite element methods. Our research in spacetime meshing makes possible much more efficient and accurate simulations for a variety of physical phenomena of interest to scientists and engineers, such as wave propagation in materials and fluid dynamics.
• Designed algorithms for problems in computational topology, such as the homotopic Fréchet distance between curves and shortest pants decompositions of surfaces.
• Designed an algorithm for capturing an object in the plane with three robots; studied motion planning and grasping problems in robotics.
• Developed algorithms for computing a minimum-cost binary search tree or BST for indexing and efficiently retrieving massive amounts of data stored on a hierarchical memory computer across multiple memory devices with vastly different access times. Our algorithms are efficient under practical constraints on the parameters of the memory hierarchy. A BST is a widely used data structure for indexing and retrieving data that was previously studied on the uniform memory model of computation. 

• Developed software tools for simulating the economics of large-scale auction-based deregulated markets for electrical power. Designed and implemented an agent-based, microeconomic, scalable model for the simulation of a deregulated electrical power market. The project included writing computer software to simulate power demand and consumption, market forces such as monopolistic and oligopolistic behavior, external regulatory policies, and physical constraints of the power grid. Unique features our simulator were individual, demographics-based elastic demand profiles; configurable matching between buyers and sellers; and flexible market clearing mechanisms. Evaluated the performance and efficiency of the market under different market clearing algorithms and sellers’ strategies.
• Designed algorithms for conflict-free communication in ad-hoc wireless networks and proved bounds on the maximum throughput of such networks. Conducted research on theoretical and practical problems associated with wireless radio networks. This work intersects the areas of graph theory, computational geometry, and networking. Our algorithms for distance-2 edge coloring and matching in graphs modeling ad-hoc wireless networks can be applied to designing routing protocols for such networks and for predicting the network throughput.
• Optimized numerical software libraries for external-memory computation to improve cache performance. Extended blocking algorithms for matrices to improve memory-efficiency of matrix operations. Implemented and simulated the same on an SGI Origin multiprocessor system, and empirically demonstrated a significant improvement in performance. Implemented a blocking algorithm for graphs, as proposed by Awerbuch et al., and investigated its performance. 

• Constructed a model in VHDL for a Fast Ethernet switch and performed system-level simulation and analysis using VHDL tools. This work, jointly with a colleague, became our undergraduate senior thesis.

Patent filings

Selected Publications

Invited Presentations 

Teaching Experience 

Service