Portfolio
Internships
- SHOPCLUES.COM, Software Engineering and Data Science Intern
- Shopclues is a leading e-commerce website in India. We were the team of three members improved search algorithm with implementing extra feature, merchant score,using machine learning algorithms with advanced cache system of solr search.
- We implemented a m*n probabilistic search over unsupervised data with making clusters.
- We improved efficiency of processing data from nodes in distributed servers
Projects
- Merchant Rating System, SPARK SHOPCLUES HACKATHON 3.0
- Created an algorithm to evaluate a single scorecard for merchant performance, quality, hygiene and business with relevant scores transparent to merchants, internal organizations in SC and buyers.
- We Won this spark Shopclues 3.0 Hackathon
- We implemented a m*n probabilistic search over unsupervised data with making clusters.
- Image Analyzing Tool, Prof. S Garg IIT Roorkee
- Developed Deep Learning based Image analyzing tool using PyTorch where User can load a large set of images to it, then they can analyze, label, cluster them using different object detection modules.
- Built and hosted model on AWS for visualization through D3.js, AngularJS, Python Flask.
- Achieved 0.98% accuracy on Imagenet dataset.
- ProSearch, Rajasthan Hackathon 3.0
- Designed and developed ProSearch Android App that is an assisted reading application, that helps users in their understanding of the material.
- Built Chatbot using Seq2seq model for question generation and answering.
- PayBhama
- This was an Android application which we made during Rajasthan Hackathon 2.0 in Kota for digitalizing India. In rural areas, women don’t have a smartphone so how they can become cashless. We made a payment service where payment model is M2C so that in oppose, women just have to carry their Bhamashah card and can pay for their commodities.
Application carries these techs:
- Material rich UI and UX
- OAuth2 based multiple handshake security
- Act locally, sync globally concept was initialized
- Firebase integration for real-time cross-check scenarios
- Autoencoder algorithm to find fraud payment or access with high-end compression and low extraction model.
- Hand Written Recognition
- Working on an Android application to convert hand written texts to limited ASCII characters; TensorFlow for training CNN .
- Applying pixel co-ordinate transformation, region-based and edge detection segmentation for images preprocessing using OpenCV.
- Obtaining 68% accuracy for alphabets written in normal spacing, unaffected by angle of view.
Open Source Contribution
- Openfoodfact Android app
- Implemented Barcode scanner to continuously scan for barcodes of products. Ported existing project in MVP framework with Retrofit 2.0.
- Enabled Offline support capabilities with BroadcastReceiver service. Added SQLite database to support offline data storing.
- Some Contribution to the Other Organisation