O. Felix Amoruwa

Cloud Surf Inc

In Development.

Team Members: TBD

Find Out More

OpenSeat Technologies

In Development.

Team Members: TBD

Find Out More

Social C

Beta

Team Members: Team NSC

Find Out More

In Development

In Development.

Team Members: Market X

Find Out More

Codename Alpha: Mobile App (Summer 2015)

Codename Alpha is a P2P e­commerce mobile application. The use case applied here is to offer culinary food from around the globe platform for sale & purchase. Certified Professional Culinary Chefs from the world's top culinary schools will be able to provide listing available for purchase on the platform, while consumers will be able to search their choice in their locale for purchase and delivery.

Tools Used: MySQL, SQLite, Java, Google OAuth API, Facebook OAuth API

Team Members: Team Delos

Find Out More

Web App (Summer 2015)

An online marketplace and delivery service for home cooked food. Instead of having to contend with commercially prepared food and the same menus day in and day out, explore the joy of discovering authentic and healthy food from real people, straight from their home kitchens..

Tools Used: PHP, Java, jQuery, LAMP Stack

Team Members: Team of 4 from CMU

Find Out More

BMW iPhone Watch App: Mobile Application (Spring 2015)

An Apple Watch application designed to serve as an interface between the vehicle and the driver for various types of human-computer interactions with purpose.

Tools Used: Balsamiq 3, Pixate, Apple iWatch UI Kit

Team Members: Team of 4 from CMU.

Find Out More

Walmart Sales Revenue Forecasting: Web App (Spring 2015)

A survey of time series and machine learning methods is presented and applied to predict the future values of weekly demand of specific store and department of a departmental retail chain. The effect of influencing parameters in weekly Consumer Price Index (CPI), holidays, unemployment, temperature, fuel price, and markdowns is considered. In setting up for evaluating the methods, candidate design and implementation of a data mining workflow is discussed. The performance of the methods is evaluated by comparing the results on the method based model on the Kaggle test data set and it’s Weighted Mean Absolute Error (WMAE) with an emphasis on the holiday and markdown features. The comparison shows that ensemble learning with combination of clustering and support vector based regression yields better prediction quality as measured by the WMAE.

Tools Used: R, SAS, Weka, Java, jQuery, MySQL

Team Members: Team of 6 from CMU

Find Out More

Mobilytics: Web App (2014-2015)

Dynamic Pricing Plan Search & Real-Time Mobile/Cloud Usage Analytics & Forecasting.

Tools Used: PHP, SAS, Java, jQuery, MySQL, Google Maps API, LAMP Stack

Team Members: Team of 2

Find Out More

Department of Energy: DE-FOA-0000649 (Improving the Accuracy of Solar Forecasting) - 2011-2012

Delivered a comprehensive energy proposal for DE-FOA-0000649 (Improving the Accuracy of Solar Forecasting): Finalist for $8M grant proposal with the US Department of Energy

ABSTRACT:

With majority of each of the 50 states in the United States having respective RPS (Renewable Portfolio Standards) requirements, our forecasted portfolio projections meet each RPS requirements over a 20 year time horizon from 2009 to 2030. By identifying a forecasted projection of United States energy consumption, the annualized life-time costs associated with constructing and operating photovoltaic grids/panels to meet each state’s RPS requirements is depicted as follows: By dividing our project plan timeline for each renewable resource into yearly phases, namely, 1, 2, 5, 10, and 20 year time frames, will assist in promoting a seamless transition and adoption of renewable energy resource photovoltaic grids/panels across all sectors in regions across the United States. Although each state in the United States has its own respective RPS standard, by regionalizing the states within the nation, project planning and implementation across the initially proposed yearly phases will promote a national rollout plan over the 20 year time frame so as to meet the requirements for each region and state respectively on average to meet the national energy goal. Segmenting the solar forecasting process into procedural steps, and leveraging this procedure with a Seasonal ARIMA model, namely the three-step ARMA model process - Identify, Estimate, and Forecast - , will not only provide a seamless and robust method in enhancing solar accuracy, but will also present an opportunity to troubleshoot the process to identify specific errors in the methodology. In the Identify step of the ARMA model, we will calculate different statistics to determine if the series is weakly stationary. Thereafter, we will calculate other statistics to help determine the appropriate ARMA model. For the Estimate step, we will estimate the model’s parameters, calculate the different statistics to help judge the model’s fit i.e. the AIC and SBC, and finally, calculate different statistics from the residuals to determine if they are approximately white noise. For the final Forecast step, we will use the estimated model to generate the conditional mean for the point forecast, and the estimated variance to generate confidence intervals. In addition, we plan to leverage tooling such as SAS with Business Forecasting for Time Series Modeling, along with non-linear & linear regression modeling, computational algorithmic programming, artificial intelligence heuristics to simulate the effects of cloud cover on intensity of Kw/h retention of solar grids/panels, the effects of seasonality, and multivariate ARIMA models. Based on GHG renewable and non-renewable energy portfolio forecasts that my team has developed previously, we have a forecasted peak cost of photovoltaic output in years 2015, 2018, and 2021, and thus propose 5 and 10 year rollout plans for the implementation of proposed 100MW solar panel grids regionally across the US. We also anticipate the collaboration between NASA, with its SORCE satellite which provides measurements of solar irradiance, and NOAA.gov, with its archive of weather patterns, which will prove useful in developing historical forecasts for future weather forecasts that we anticipate to use in the development of accuracy-enhanced solar forecasts. Additionally, we also plan to develop an operating policy that will align with the solar metrics and the enhanced accuracy of solar forecasts that models the optimal operating conditions with regard to positioning and operating the photovoltaic arrays/grids via a latitudinal and longitudinal coordinate system, which can be simulated via AIMMS/MOPTA modeling. The operating model takes into account that some days may operate at a loss due to foul weather, but overall, we believe that the operating plan offers the best plan of action pertaining to positioning the arrays based on the solar forecasts in order to maximize direct and diffuse solar energy absorption.

Team Members: Team from Berkeley/CMU

Find Out More by Requesting for the Proposal

Department of Energy: DE-FOA-0000740 (SEEDS - Solar Energy Evolution and Diffusion Studies) - 2011-2012

Delivered a comprehensive energy proposal for DE-0000740 - SEEDS

ABSTRACT:

With regards to the conceptual idea of the SEEDS (Solar Energy Evolution and Diffusion Studies) project, our preliminary research in participating in this FOA will involve balancing a targeted Six-Sigma DMAIC approach, with detailed data and quantitative analysis associated with the focal areas chosen. The DMAIC process is used for improving, optimizing and stabilizing business process and designs and can be used as the framework for other improvement applications. We anticipate sharing our quarterly results with the DOE and the SunShot group through regularly scheduled quarterly meetings. The organizational structure entails three groups focusing on Solar Adoption Ideas, Solar Adoption Products, and Solar Adoption Technologies, independently, with cross-collaboration, overseen by the Management team. In each aspect of our focal points for this project, we anticipate providing detailed answers to Innovation Diffusion questions related to how, why, and at what rate each of the focal points spread through society using large data sets and computational tools for analysis. To model, simulate, and forecast the adoption of each of the three focal points once we have gathered the relevant data from each, by leveraging these procedures with a Seasonal ARIMA model, namely the three-step ARMA model process - Identify, Estimate, and Forecast - , KNN (k-nearest neighbors), and ANN (artificial neural network) models, through triangulation, will not only provide a seamless and robust method in understanding the trends that our economy has encountered in the past with respect to Solar Adoption, but will also represent an opportunity to investigate future trends, key drivers and levers which correspond to positive and negative trends in the industry. In addition, we plan to leverage tooling such as simulation software to model market & evolution dynamics, consumer decision-making, social, behavioral and economic barriers and parameters, additional tooling such as SAS with Business Forecasting for Time Series Modeling, along with non-linear & linear regression modeling, computational algorithmic programming, artificial intelligence heuristics to simulate and examine the diffusion challenges, the associated effects of seasonality, and multivariate ARIMA models. For the data management and quantitative analysis component, we plan to establish a client server scenario using J2EE with web service container(s) hosted on a cloud platform with load balancing and failover protection to enable the software as a service platform in managing the large data-sets we anticipate collecting. The client would be developed in multiple flavors on an independent executable interface application. In evaluation of the infrastructure technology, we are proposing Apache Hadoop, which supports data- intensive distributed computing for various computational models, with an Application Server (J2EE), which is an open standard platform that will allow seamless data migrations, and a supporting database.

Team Members: Team from Berkeley/CMU

Find Out More by Requesting for the Proposal

Toigo Private Equity Fellowship - First Look (Fall 2011)

Met with professionals from the participating firms, including Apollo Global Management, The Carlyle Group and Vista Equity Partners, as well as Toigo Alumni working in private equity at:

NASDAQ, 4 Times Square, New York, NY 10036.

Pitched PE Acquisition #1 - Calpine:

Calpine is an independent wholesale power company in the United States who owns and operates primarily natural gas-fired and geothermal power plants in North America.

Pitched PE Acquisition #2 - Big Lots Inc:

Big Lots Inc is a a discount retailer for general merchandise and close-out liquidation sales.

Find Out More

ExxonMobil & Executive Leadership Council (Spring 2011)

Led a team of 5 through the development to a $35,000 scholarship:

1) A projection of what the U.S. energy portfolio would look like by the year 2030 considering current and potential innovations with an emphasis on "cleaner" fuels and technologies.

2) A strategic fiscal plan that will help the nation meet its energy goals while accounting for private, public and international sector investments to identify potential savings created by generating energy more efficiently and transitioning to "clean" fuel sources.

3) A human resources forecast highlighting key areas of job growth and development for the industry to help meet one of the energy plan's key objectives.

4) A recruitment and development plan to assist the energy industry in meeting its needs more qualified candidates, with specific action for recruiting more African Americans and ethnic minorities to work in the industry.

Team Members: Team of 5 from CMU

Find Out More

Using Artificial Neural Networks to Predict Secondary Structure of Proteins (Spring 2005)

Our final project performs a secondary structure prediction on primary protein sequences by using an artificial neural network model that implements a learning algorithm. Using the primary protein sequences and the corresponding secondary structure sequence as input, upon training an artificial neural network (ANN) with approximately 42 different proteins, our ANN was able to predict the secondary sequence of any given primary protein sequence above 50% accuracy.

Tools Used: C++, Java, RHEL, Unit-ProtKB/Swiss-Prot Database

Team Members: Team of 2 from Berkeley

Find Out More