We're building a next-gen AB experimentation platform for modern data and product teams to run 10x more experiments with existing infrastructure and headcount.

Took a year to work off the travel bug from years at Airbnb. Read about it at our travel blog!https://subuandchetravel.wordpress.com/

As one of the first few data scientists at Airbnb, I worked to use data to serve the Airbnb community. This included:- Developing fraud prediction machine learning models from scratch, starting from zero data to a system that could train and maintain multiple models in production. - Building out Airbnb's original ETL for all fraud detection and mobile platform pipelines- Leading a team of five data scientists to develop a novel knowledge management and sharing platform, now open sourced and used by a range of industries including pharma, market research, and on demand services. https://github.com/airbnb/knowledge-repo- Providing data systems guidance on a new logging enforcement mechanism to automatically catch data bugs before they hit production.- Helping to establish a culture of experimentation by running hundreds of A/B experiments, ranging from simple splits with fast turnarounds to long term treatments split by country pairs.- Rapid prototyping of dashboards to bring transparency of Airbnb's success across every mobile device, OS, and browser

Statistical analysis for empirical support of healthcare policy. Write code to harness medicare claims data to help formulate Pay for Performance policy initiatives as mandated by Patient Protection and Affordable Care Act.

Used MEG data to test for coherence between brain hemispheres during a visual task using statistics and signal processing methods. Tested for appropriate windowing method using morlet wavelets, hanning window, and multitaper. Used permutation tests for coherence and phase locking value. Statistical tests done using matlab with fieldtrip package.

I worked on the optimization team, where I was improving the inference model used to prediction demographic segments for targeted ads. I tried different regularization methods to deal with problems of multicollinearity with new datasets and features that the company obtained. I implemented combinations of L1 and L2 penalties to the regression model, testing different methods for validation to find the optimal penalty weighting. Lastly, I investigated the effects of regressing on arity in different datasets.I used Quantcast's retooled Hadoop and MapReduce infrastructure to pull data, and made an R script to automate my data experiments to easily extend to new datasets as they arrive.

I implemented transformation algorithms for cell phone cameras to correct for misalignments and aberrations in images of UV-lit barcodes. This required that the corrected image not blur or cause arbitrary high frequency components that would interfere with the barcode reading. I used a bicubic interpolation scheme to reduce high frequency artifacts, and then used an unsharp mask to enhance the edges of the barcode. I then used the transformation to build from scratch a paper authentication algorithm to encode and decode the paper fiber pattern. I used the relational data of blocks of an image in the real and frequency domain to find a unique key for the original document. This data was then encoded in a UV ink barcode, providing a link between the content of a document and the paper medium on which it is printed.

-Developed a photo-diode H-bridge heliostat (solar tracker) for solar syringe melter for rural hospitals.-Prepared energy-cost model for resort seeking a renewable energy solution