Developed pipeline for qualifying, filtering, and cleaning patient data & assembly into table for causal analysis. Developed and applied causal analysis procedure. 503 causal relations were found from a network of 11790 variables.Independently researched/studied needed statistics to devise best approach for unique (longitudinal, hole-y, unequal spaced time intervals, per-person) data. Surveyed ~50 papers.Parallelized running of algorithm on Stanford's compute cluster to reduce time 8-fold, saving about 20 days.Merged 6 data sets (~900 rows and 100-10,000 columns each) together to create final table which could be passed into the algorithm for analysis.Scrutinized all data sets (~15) to eliminate redundancies (don't want to over-weight data). Eliminated 4 sources of redundancies.Optimized 2 hyperparameters.Filtered/qualified data set resulting in a 42% usage rate making numerous judgment calls e.g.(1) variables containing redundant ethnicity information,(2) what threshold to use for deeming noise(3) throwing out columns with too many missing values (balance tossing columns with the bad juju of imputing values.)Discovered bug in R package used, and developed workarounds when the package did not provide the needed functionality.At each step of the pipeline, performed sanity checks to ensure results were sensible.*In at least one case this did reveal a flaw in the paradigm, which resulted in a refinement with greater correctness.*To achieve this, numerous test data or random data was carefully generated and applied.Kept "digital lab notebook" documenting all justifications for design decisions with dates.Included test unit code (and inputs) where appropriate.~200 pages.Created SQL script to merge 9 tables together and annotate, to provide a final database that is easy to peruse. Saves group >100 hours a year as this is the sample (blood draws, etc.) database for the iPOP program at the Snyder Lab at Stanford.

Self-study. Read textbooks on biochemistry, cell biology, neuroscience, immunology, genetics.

Supervisor: Dr. Joshua Mackie.De novo gene assembly from Illumina data and validation.

Supervisor: Dr. Yoshihisa Yamamoto.Parallelize a small C program with CUDA. 50% speedup.

Build homemade EEG.

SAMPLING OF PROJECTS:(1) Web interface to database15 person department needed creation of database, with website interface, to replace error-prone document handling system.Designed simple clean button interface in collaboration with supervisor and users; implemented in HTML/Javascript, with back-end code in PHP and SQL.Set up Apache web server and MySQL on 3 machines (for redundancy), got synchronization and automatic nightly backups going.Reduced lost task requests from 1/month to 0, saving department 1-2 hours/month.Searchable history of tasks saved department 1 hour per week.Enabled auditing of employee performance, enabling more fair employee workload and evaluation.Automatic archival and timestamping of task requests saved department 10 hours/month.(2) Real-time impact point prediction27 person department needed real-time remote display of impact point prediction (measure of "health" of the ballistic object being monitored).Wrote Win32 API C applications (packet sender and graphical display receiver).Refactored the code several times to eliminate technical debt and minimize latency, cutting length of code by 60% (to 700 lines); co-worker commented "best code ever seen."Ran software during 4 test launches streaming to 3 sites across U.S. without a hitch.(3) Strip chart27 person department needed Windows application to display running plot of incoming telemetry data (4 measurements).Wrote in object-oriented C++.Used for 8 years as backbone of status monitoring during test flights (as long as was at the company), possibly still used now.

Database management.

Web application, HTML/Javascript.

C++ programming for the RADIANCE lighting simulation software.