An independent practitioner of value investing in the Indian capital markets.Enjoy reading annual reports of public companies to understand their economics and to identify the range of its intrinsic value.July 2013-June 2024, the portfolio has generated CAGR of 28.33% (excludes dividends and taxes)An avid reader of books on investing, biographies, psychology, history, etc. https://www.goodreads.com/review/list/51519192-steven-mudda

(1) Designed and developed a model to recommend locations to users in a social network. The model was evaluated on a large scale dataset consisting 6 Million check-ins (or location visits) of 196K users at 1.2 Million locations with a social network of 950K friendship ties.Performed feature engineering to extract temporal, social and spatial location preferences for each one of the 196K users. The extracted features take into account the habits of humans like visiting nearby locations, frequent visits to a set of locations and affinity to specific locations due to friendship ties.(2) Developed a model to predict the set of most likely regions a customer is expected to go based on their mobile Call Detail Records to enable optimum allocation of network resources.Analyzed mobile Call Detail Records of 11 Million people in a city with 3K cell tower regions.Designed features to capture the mobility patterns of customers and their social ties.Developed a fully automated model test pipeline using Unix shell scripts in a cluster environment.(3) Applied PCA to reduce temperature data recorded at 9 different stations in Berlin to address the problem of weather prediction (Matlab: eig, covar). Reduced 203K records of 9-dimensional temperature data to 3-dimensions with low reproduction error.

1) European project MONARCA: Doctors get to observe the mental health of a patient only when they visit the clinic. Can doctors watch the daily moods of their patients to provide better medical care?Co-designed and developed a system that enables doctors to monitor and treat bipolar disorder patients. Co-developed an Android mobile application that examines the physical activity of patients through sensors in their smart-phone like Fitbit fitness app.Compressed vast amounts of real-time sensor data using Google Protocol Buffers and transferred it securely (AES encryption, HTTPS) to a server accessible only to their doctors.2) bCards: An application to exchange digital business cards as part of a European projectDesigned and developed an Android mobile application that automatically discovers nearby people via Bluetooth and enables users to share digital business cards (Java, Eclipse).

I worked with the Social Computing group at Microsoft Research Asia, Beijing. • Analyzed 2.3 Million visits of 325K users to learn their visiting preferences for 79K locations in the city of Beijing, China.• Designed and implemented a time-aware Deep Neural Network model that recommends new places to users at different times of the day (Theano, a Python Library).• Developed and combined two recurrent neural network (RNN) models (specifically Gated Recurrent Unit or GRU). One GRU model learns the temporal preferences of a user and the other model learns the temporal preferences of a location.

Developed a new wireless communication protocol that enables mobile sensors to discover each other automatically and exchange data seamlessly. We implemented the wireless protocol on Crossbow’s Micaz Sensor Nodes.

Worked in a startup team to develop an “Online Education Process Management (iSchool )” software to manage the processes in educational institutions of India (Apache, PHP, JavaScript).