I am a Research Associate at Carnegie Mellon University. I have tried my hand at a bunch of different domains in Computer Science. My core interest lies in Privacy, Security, Machine Learning and Distributed Systems. I aim to build products rather than just code. My latest projects have been developed with emphasis on the usability of the products. At CMU, I am working on usable privacy and Machine Learning.
Protect My Privacy (PMP) lets you protect you protect the personal information on your rooted Android device. It provides a layer of security between apps and the operating system, thereby giving the control back to the user. The source of request for private inforamtion is dynamically identified and the choice is given to the user to grant the application access, or access to fake information using techniques such as Reflection and Callbacks. PmP also has a Firewall feature, which blocks the WiFi and mobile data on a per app basis. PmP has over 30k unique installs.
Python and Django based Backend
Machine Learning based Recommendation Engine
Privacy Proxy detects the Personally Identifiable Information (PII) from an Android device. It runs as an in-situ VPN Service and collects the Network Signatures of the crowd (even for SSL traffic). These signatues are compared across the signatures of other users using the same app. This enables PrivacyProxy to detect unique PII. PrivacyProxy also has a MachineLearning layer which detects the purpose of the PII detected using the dataset of PrivacyGuard.org. Finally, PrivacyProxy intelligently filters the PII from your network traffic using the purpose of the PII detected. PrivacyProxy paper has been submitted for Mobisys 2017.
Crowdsourced PII detection
Machine Learning for purpose detection
Network Security and Privacy
'Generist' detects the genre of a music file via Machine Learning. The machine learning model was trained on a range of different music files. The audio files were transformed using Fast Fourier Transform and Mel-frequency cepstrum coefficients. Generist has a website based frontend, where the users could upload the audio files. The accuracy of Generist was nearly 80%.
Data transmission and Storage
Wean Hall 4311,
Carnegie Mellon University
5000, Forbes Avenue
Pittsburgh, PA, USA.