Clinic Projects

Please click on a link below to view the Harvey Mudd College Computer Science Clinic projects for the corresponding time period.

Clinic Projects for 2014-2015

Merchant Recommendation Systems (CS/MATH)

Client
American Express

Faculty Advisor
Professor Elizabeth Sweedyk

Student Team
Arthur Chi, Corey Hayes, Billy Mills, Tongjia Shi (PM-S), Mathew Wilber (PM-F)

American Express provides financial services to much of the United States and world populations and maintains a large network of cardholder transaction data. This gives American Express the potential to make high quality recommendations to cardholders, suggesting small business they would enjoy but have yet to discover. The Clinic project aims to explore a variety of recommendation algorithms to improve upon the performance and scalability of the current American Express recommendation engine.

Mapping for Microscopes: Automatic Processing of Apatite Images

Client
Apatite to Zircon, Inc.

Faculty Advisor
Professor Zachary Dodds

Student Team
Kaya Woodall (PM), Justin Jones, Kathleen Schaefer, Richard J.T. Booth

Apatite to Zircon is a geology company that specializes in sediment analysis for determining a region's geological history. A majority of the geologists' time is spent at the microscope, manually searching for crystals to be analyzed. To increase the flexibility of this process, we implemented a fully-automated system that captures in-focus images across an entire slide, which we then provide in a seamless, high-definition viewing environment. In addition, we also generate locations of suggested crystals of interest.

Design and Implementation of a Next-Generation Software System for Non-Human Primate Testing

Client
Dart Neuroscience, LLC

Faculty Advisor
Professor Ben Wiedermann

Student Team
Mary Elise Elam (PM-S), Ari Hausman-Cohen (PM-F), Marjorie Principato, Alexander Swafford

The DNS/HMC clinic team is creating testing software to be used by Dart NeuroScience for drug discovery. The project bridges the newly created hardware and the researchers' language more effectively than the current system. It allows biologists to use their terminology to create new tests without loss of intellectual property. It also provides valuable feedback and flexibility to the researchers that is not present in their current system.

Geographic Relevance for Travel Search

Client
Expedia, Inc.

Faculty Advisor
Professor Robert Keller

Student Team
Chris Brown, Benjamin Leader, Hannah Long (PM), Nabil Zaman

The online travel industry faces challenges in providing relevant information to travelers searching in an exploratory manner. Many travelers would like to explore possible destinations based on qualitative features such as interesting history or great museums. The purpose of this project is to help Expedia respond to these qualitative search queries by designing a demonstrably effective method to associate hotels and regions around the world with the higher-level attributes that describe them, such as "family-friendly," "cultural," or "romantic."

PC-Based Home Automation

Client
Intel Corporation

Faculty Advisor
Professor Ran Libeskind-Hadas

Student Team
Johnathan Ashley, James Bowen, Najla Bulous (PM-F), Vanessa Ronan (PM-S

The Intel Clinic team is developing a system to help homeowners decrease their energy usage in accordance with their own goals. The system tracks the energy consumption of electric devices in a user's home and uses machine learning methods to recommend energy saving practices. Once accepted by the user, these recommendations are implemented using a home automation system.

Software-Defined Network Taps

Client
Ixia

Faculty Advisor
Professor Geoff Kuenning

Student Team
Bryan Trujillo (PM), John Phillpot, Mari Bennett, Sam Schumer

Ixia is interested in extending the reach of their solutions to cover new type of network that is "software-defined." A software-defined network (SDN) allows for easier configuration and maintenance, as every device in the network is controlled centrally. Our application will serve as a tap for SDNs, allowing the user to process, duplicate, and redirect their network traffic more easily.

Conduit: Scientific Data Exchange for HPC Simulations

Client
Lawrence Livermore National Laboratory

Faculty Advisor
Professor Robert Keller

Student Team
George Aspesi, Justin Bai, Rupert Deese (PM), Linnea Shin

Conduit is a new open-source library for high performance computing (HPC) applications, developed at LLNL. It provides a C++ interface for describing and accessing complex in-core data. Our team tested and improved Conduit to make it more appealing for adoption within the HPC community. We extended Conduit's capabilities by prototyping an I/O library, Message Passing Interface (MPI) wrappers, and a visualization tool. We also demonstrated the use of Conduit for aggregating performance data for MPI programs.

Improving LinkedIn's Graph Software Load Balancer

Client
LinkedIn

Faculty Advisor
Professor Ran Libeskind-Hadas

Student Team
Eoin Nugent (PM-F), Brian Leonard, Christine Schubert (PM-S), Helen Woodward

LinkedIn is a leading professional networking service with over 300 million users. It is important for a large social network such as LinkedIn to balance loads efficiently in order to provide the best user experience. This Clinic project has improved Norbert, the load balancer used by the social graph team. Norbert is an open source project implemented in both Java and Scala by LinkedIn developers. Norbert provides a simple API to deal with cluster management and client/server networking.

Quantifying Latent Fingerprint Quality (CS/MATH)

Client
MITRE Corporation

Faculty Advisor
Professor Yi-Chieh Wu

Student Team
Martin Loncaric (PM-S), Sarah Scheffler (PM-F), Jordan Varney, Christopher Eriksen

The MITRE Corporation has created test methods and fingerprint image quality metrics that automated fingerprint identification systems rely on for function and interoperability. Our team is producing methods to extend MITRE's system to measure the quality of unintentionally deposited fingerprints, which may be too smudged or incomplete for use. Using image processing and statistical techniques, our model will help determine the strength of evidence provided by a fingerprint image.

Gaming REality: Real-World 3D Models in Interactive Media

Client
Matterport

Faculty Advisor
Professor Zachary Dodds

Student Team
Kevin Choi (PM), Sisi Cheng, Emma Davis, Noelle Fa-Kaji, Alden Weaver

Matterport is interested in expanding the consumer usage of their 3D models of interior spaces. To that end, our goal was to create a product which demonstrates the strength of Matterport models in a way that highlights possible applications using consumer-generated models. The end product is an iOS game that is geared towards incoming Harvey Mudd first-years and uses models of real interior spaces at Harvey Mudd.

Snapdragon-Based Robot Development

Client
QUALCOMM Incorporated

Faculty Advisor
Professor Jim Boerkoel

Student Team
Taylor Peterson (PM), Sean Messenger, Jandro Alderman, Andrew Michaud

Qualcomm Technologies, Inc. believes that the Snapdragon 600 system-on-a-chip (SOC) has many potential uses in mobile robotics. The IFC6410 is a single-board computer utilizing this SOC that can fit on most robotic platforms. This project will use the IFC6410 as the brains of a TurtleBot, enabling it to localize in a known environment, navigate between goal locations, and provide tour guide functionality. In doing so, we will demonstrate the capabilities of the Snapdragon 600 in a robotics context.

Understanding and Preventing Threat Through Security Data Analysis (CS/MATH)

Client
Rapid7, Inc.

Faculty Advisor
Professor Lisette de Pillis

Student Team
Will Clausen, Abhishek Goenka, Huameng (Michael) Jiang, Arianna Perkins (PM), Xinlei (Mimee) Xu

Given security data collected in the duration of an attack, the team will structure the data, explore the data, evaluate the possibility of determining if and when an attack occurs, and attempt to develop algorithms for detecting relevant patterns in real-time. If successful, the team may work toward developing a package for analyzing security data for Rapid7.

Achieving Distributed Point-in-Time Consistency for Geo-Replication

Client
Red Hat

Faculty Advisor
Professor Beth Trushkowsky

Student Team
Michael Staffron (PM), Nick Carter, Matt Cook, Philip Davis

Red Hat's Ceph is an open source, hyper-scalable, distributed, strongly consistent file system. Red Hat is interested in supporting geo-replication of data in a Ceph instance, which requires the ability to take consistent, point-in-time, online snapshots of the complete distributed system. We propose and analyze an algorithm that uses time synchronization and transient write delays to implement this snapshotting feature while maintaining Ceph's consistency and performance guarantees.

Real-Time Query and Visualization of Large-Scale Data Streams

Client
Rubicon Project - Query

Faculty Advisor
Professor Chris Stone

Student Team
Victor Bhattacharyya (PM-S), Bruce Yan (PM-F), Isabella Funke, Xiaofan Fang

The Rubicon Project Clinic team is working to build a large-scale data pipeline that can ingest data in many formats so that the data can be queried and visualized in real-time. Users can ask for information about the data, and visualizations for user queries should update in real-time, and be available on a dashboard that loads in under ten seconds. Additionally, the pipeline should be able to handle one billion events per day.

Real-Time Data Aggregation

Client
Rubicon Project - Stream

Faculty Advisor
Professor Chris Stone

Student Team
Eli Gadd (PM-F), Alex Melville (PM-S), Colin Stanfill, Jesse Watts-Russell

Rubicon Project operates a fully automated trading platform for the buying and selling of online advertisements all over the world. This platform generates a tremendous amount of data which takes up to an hour to aggregate in a data center before it can be analyzed. We are designing a real-time streaming system that will serve as a proof of concept to reduce the latency to as low as a minute.

Reengineering HPC Kernels for Manycore Architectures

Client
Sandia National Laboratories

Faculty Advisor
Professor Jeff Amelang

Student Team
Brett Collins (PM), Alex Gruver, Ellen Hui, Tyler Marklyn

We worked with Sandia National Labs to explore the performance of various strategies for implementing thread parallel computational kernels on the CPU and GPU. Based on our research and performance analysis we developed recommendations for maximizing the performance of these kernels. We also provided data about the usability and performance of their parallel computing library, Kokkos. These results will help Sandia expand the usage of Kokkos both externally and throughout their code base.

Process Time Analysis in Spaceflight

Client
SpaceX

Faculty Advisor
Professor Ben Widermann

Student Team
Wendy Brooks (PM-F), May Lynn Forssen (PM-S), Alix Joe, Rachel Macfarlane

SpaceX designs, manufactures, and launches rockets. Identifying problems in the software on these rockets is often slow. SpaceX engineers currently use an old tool to help find these problems. We received feedback from our liaisons and other SpaceX engineers. This helped us to recreate the tool with new features that improve the user experience.

Veneer Color Classification

Client
Steelcase, Inc.

Faculty Advisor
Professor Jim Boerkoel

Student Team
Michelle Chesley, Coline Devin (PM-F), Andrew Donelick (PM-S), Wai Sing Wong

The goal of the project is to develop an automated process for determining the grade and color profile of various wood raw materials. This process includes classifying the base color, evaluating whether it meets desired thresholds for particular project needs, and then indexing and archiving a visualization of the veneer for later use and quick retrieval.

Image Tagging Platform

Client
Time, Inc.

Faculty Advisor
Professor Geoff Kuenning

Student Team
Emily Blatter, Nicole Change (PM-S), Daisy Hernandez (PM-F), Henry Tay, Stephanie Zellner

Time, Inc. is the largest magazine media company in the U.S., owning brands like People, Time, and Sports Illustrated. Their magazine editors receive over 200,000 photos in a very small window of time from large events like the Oscars and the Super Bowl. Our goal was to make the image search and selection processes more efficient by creating modules that add metadata to each photo that indicate image quality and the presence of significant features.

Design and Analysis of RNA Molecules with Diverse Folding Pathways

Client
University of British Columbia

Faculty Advisor
Professor Elizabeth Sweedyk

Student Team
Rachel Sherman (PM), Kennedy Agwamba (F), Sasha Heinen, Carson Ramsden

Artificially designed nucleic acid molecules with diverse folding pathways have many potential uses, from performing computations with biological material, to having therapeutic applications. However, designing molecules that behave predictably is challenging. Building on software that simulates a molecule folding, our software outputs statistics and graphics about potentially interesting structures in a molecule's folding pathway. Researchers can use this software to test and modify new designs until they fold as desired. Additionally, we have analyzed a theorized molecule and proposed modifications to its structure.

Detecting Location Intent From Context

Client
Yelp, Inc.

Faculty Advisor
Professor Julie Medero

Student Team
Ethan Kenny (PM-S), Viona Lam (PM-F), Henry Huang, Angela Zhou

Currently a Yelp search is split into two fields: the query and the location. Our goal is to create a prototype that returns more relevant results to the user by better identifying and understanding locational information in the query field. In addition, we outline how our system could be incorporated into Yelp at a national level and ultimately be used to implement a single field search.