Research
Deep Residual Networks for Plant Identification
-
Found a novel approach in residual networks by skipping three layers instead of two.
-
The give model was tested on the LeafSnap dataset consisting of different 185 plant species.
-
Achieved 93.09% accuracy on testing dataset, illustrating that deep learning is a highly promising forestry technology.
Projects
Code2Create WebApp
-
Developed a database in MongoDB for a hackathon, integrated authentication through Gmail/Facebook. Also developed back-end for live quiz, and added additional features allowing admins to send invites to other users and form teams
Robo Path
-
Implemented an intelligent agent that can perform BFS, DFS, UCS and A-star to find the most optimal shortest path from one location to another.
All About VIT
-
Reduced workload of students by 90% by developing a website in Django where students can see hostel rooms online and can engage with each other to ask and answer questions about a variety of academic topics
VBOT
-
Developed a chatbot for college students (as dynamic as possible) which conquer general queries related to VIT.
-
Provide ideas on how to clear backlogs/arrears, by looking at the record of students who got backlogs and how they had passed those exams.
- Used Convolutional Neural Networks, Natural Language Processing, RASA Framework etc.
Emotional AI
-
Developed a model using CNN and RNN to predict user’s emotion from facial expressions and text data with 90% confidenc
- Established real-time counseling solutions to users by building an end-to-end application using Flask and JavaScript.
VIT Cab Share
-
Revamped entire database design to improve performance by 40% by migrating from PHP to NodeJS
Classifying Garbage using CNN
- Augmentation : As the data size was not sufficient, augmented techniques have been applied to increase the dataset size.
- Dataset : Converted dataset into into binary files(HDF5 dataset) as it takes very less time in processing.
- Web Sockets : Implemented socket programming which enables us to work in real time environment.
- Accuracy : Model trained on GPU (Google Colab) and achieved an 97.4% accuracy on training dataset and 93.2% on testing dataset
Movies Prediction
- A python script that looks up for all movies available on my desktop and extract plot, ratings and genre.
- Used web scraping, sentimental analysis, data cleaning, data analysis and machine learning concepts.
- Output is shown by with plotting graphs using R and Python and providing excel file listing all the movie features.
