Machine Learning

Natural language question answering system ( report )

• (Natural language question) -> [ Factoid-based question answering system ] -> (paragraph containing the answer)
  
• Example:
  Q: Who was named the 2009 Nobel Peace Prize laureate?
  A: Nine months after his inauguration, Obama was named the 2009 Nobel Peace Prize laureate.
  
• Used
  • NLTK (for parse-tree, PoS Tagging and Named Entity Recognition)
  • Hadoop MapReduce + Pig for querying Wikipedia dump and ranking of search results
  
  

Study of Gaussian Processes (nonparametric models) for regression and classification

• Studied and coded in Python to better understand the machinery of Gaussian Processes
  

Acronym Disambiguation (ongoing | code )

• Designed a system which accepts a document and expands the acronyms in it. (similar to word-sense disambiguation)
  
• The system performs near state-of-the-art on the MSH Corpus
  

Identifying Degree of Diabetic Retinopathy (report)

• Diabetic patients start losing eyesight in severe cases. We designed and developed a system which on getting the left and right eye image of the diabetic patient tells the amount of damage in the retina.
• The system used image processing, class balancing and deep neural networks to work.
• The model worked with 78% accuracy.
  

Learning summarization from Wikipedia ( code | report )

• Devised a new unsupervised graph based approach to measure sentence similarity using Word2Vec
  
• Used vectorized form of current sentence, nearby sentences and LDA vectors of paragraph and article to decide whether current sentence should be included in an auto generated summary.

Software Engineering

(Recommendations, on LinkedIn)

System Software Engineer at Hewlett Packard R&D ( Bangalore, 2011 - 2014 )

• Part of the Printers and Personal Systems R&D group at Hewlett Packard.
• Our team extended OXPd ( a printer SDK ) to a new range of devices, enabling HP to have the largest range of printers enabled by an SDK.
  
• Responsibilities:
  • I handled the design and implementation of the page scanning API for the C# SDK.
  • HTTP based mocker for any RESTful web service : A web service to mock a printer's REST API in any state. Removed dependency of SDK team on firmware implementation. Helped meet SDK team deadlines within time.
  • Windows installer for the SDK: Completed implementation of a windows installer for deploying the SDK on enterprise servers.
  • Automated Build on git push: Used MSBuild to verify SDK against a virtual printer
• Awards:
  • Highest year-end rating for my work.
  • Four awards for product-wide enhancements and quality of deliverables

Intern at eBay India Development Center (Chennai, 2011)

• Refactoring of the code base for GivingWorks vertical inside eBay to isolate as a Java API. Later, expose the API as a SOAP API internally within eBay.

Student at VIT University (Vellore, 2007 - 2011)

• "VITSat 1" Picosat project:
  • VIT University's attempt at making a pico-satellite.
  • We worked as a team on designing the hardware architecture, kernel and batch scheduler for the satellite.
  • The design at that stage was approved by Indian Space Research Organization and the high level of robustness was appreciated.
• Developing a Grid Computing Network using Java RMI: 
  • Developed a 5 node grid which shared computing power using Java RMI.
  • Primary intent was to get a hands-on experience of developing and operating a grid.
  • We also learnt the effects of network latency and job size on total gain in throughput.
• Computational grid using Process Virtual Machine:
  • Used Process Virtual Machine (PVM) to build a heterogeneous computational grid consisting of Windows and Linux hosts
• Virtual Storage Manager for Windows: 
  • This was an attempt to create a distributed versioning system for managing documents in enterprises.
  • A prototype was developed and demonstrated, but further development was abandoned after I became aware of Source Code Management systems which did the exact same thing.