× Undergraduate Thesis Eyelog Clients Contact

Undergraduate Thesis : NLP Advances for Marathi 2019

The aim of this undertaking is to boost the advancement in language technologies for the regional language- Marathi, to increase its comparability to more advanced languages such as English. Named Entity Recognition or NER in a resource-poor language opens up new dimensions for researchers, and helps increase industrial opportunities. Being a research-oriented project, NER for Marathi Language provides preprocessing tools for handling the Devanagari scripts and aims to apply custom deep learning architectures for the first time to Marathi prose data to generate state-of-the-art language models for the Marathi language. To achieve these goals, the project involves the generation of multiple benchmark databases, along with the implementation of traditional methods to provide a reliable comparison of results. We further integrate generated resources with our custom language model to provide a News Classification System for Marathi headlines to demonstrate the various use cases of NLP in the world. This model is further manifested into an android application for convenient corroboration and with OCR functionality for enhanced user experience. This project is motivated by a single motto- ‘English language should be a choice, not a necessity’.

Team : Smridhi Arya, Shivani Deshmukh
Under guidance of Dr. Pravin Shrinath

  • 1 Android App
  • 3 Benchmark Corpora
  • 2 State-of-the-Arts
  • 3 NLP Tasks
  • 100% User-Friendly

EyeLog : Object Recognition for the Visually Impaired 2017

Achievement Unlocked:
Runner-Up in NMIMS University's App Development Competition.

This was my debut project in the field of Software Engineering. From being acquainted with the domain of Computer Vision, to learning Android Studio from scratch - this smartphone application was developed as an aid for the visually impaired. It has integrated Text-to-Speech features that link to the individual's real-time video stream being received by the camera, thus enabling the person to point and identify. With minimal user interaction, this application was made keeping complete concerns of the target audience in mind. The model was trained using CNNs on the MS-COCO dataset combined with custom image data prepared from images of the university campus.

Team : Smridhi Arya, Shilpika Agarwal

A Few Projects Worth a Mention

  • Bi-Verify Devised algorithm for
    Emotion Detection
  • Chatbot for Suicide Prevention
  • CapsNet for Semantic Segmentation