S V Praveen

VENDO - DECENTRALIZED ADS

COMMUNITY PRIZE 2021

• Authentic traffic for advertisers, Profitable ad-viewing for web-surfers.
• Our solution incentivizes viewers to view ads by offering them Vendos (VEN, ERC20) for verifying their impressions. Impressions are verified by asking viewers to fill out a captcha "related to the company's tagline", thus, making impressions really count.
• Decentralization by shifting the workflow to blockchain enables everyone to benefit from increased transparency as all impressions that are stored on the chain can be verified by the anyone including the publisher and marketer.

CODE   SUBMISSION

COVID-19 PREDICTIVE MODELLING

TOP 25 in SAMHAR COVID-19 HACKATHON 2020

• Developed a Seq2seq deep learning model to model the district wise spread of COVID-19 cases, as opposed to the standard SEIR model.
• An extensive dataset was mined and scraped from various sources to track case counts, district resources, census and news.
• Our window-based model accuracy was significantly improved by feature selection, hyperparameter tuning and detrending the data. Predictions for short windows were very close to actual numbers seen in the next day.

CODE

Resumatic

Winning Idea of Hack IDF@India Dapp Fest 2019

A Blockchain-based decentralized app that provides a novel solution for verified credentials on person's resume.

• The project exploits a major advantage of blockchain - once a credential is verified on the blockchain ledger, it is stored as an immutable record that can never be tampered with. Hence, the tagline - Once Verified, Always Verified!
• A private cloud swarm was configured for storage and a django app that interacts with it through an API was developed.
• Security was also a forte of the solution's architecture - as everyone always held onto their data privately with them and, could provide their verified details, to hiring companies, securely by sharing a token that can be used to query hashed records stored in the ledger.
• When implemented on large-scale, this project shows good potential to save huge costs for companies in the first step of verification in their hiring process.

CODE   TWEET

Face Recognition on IMFDB

TOP 25 CDAC Nvidia Atos AI Hackathon 2019

The solution attempts to overcome previous challenges on Face Recognition on the IMFDB dataset with faces subject to variance in scale, pose, occlusion, expression, makeup and illumination.

• A pre-trained YOLOv3-416, currently one of the best object detection models, was fine-tuned on the WIDER FACE Dataset to obtain nms 0.1 with confidence threshold of 0.95.
  This model detects faces with normal, occlusion, pose and glare variation extremely well.
• For face recognition, we used FaceNet, that directly learns a mapping from face images to a compact Euclidean space where distances directly correspond to a measure of face similarity. We are able to recognize Indian Faces using Facenet with Inception resnet v1 reasonable accuracy.
• Experiments of using Deep Residual Equivariant Mapping to increase accuracy of detecting profile faces were also explored.

DEMO

Alpha Mind

A bot that wins any game of Mastermind [AKA Bulls and cows] within 7 moves.

• A modified Implementation of the popular Minimax Algorithm from Game Theory, inspired by Donald E. Knuth's - The Computer as a Mastermind
• Optimization of Algorithm in C++ for fast guessing and response.
• Also implemented an Alpha Learner to try to find patterns in opponent players and adopt a customized strategy to win against them
• Clean well structured code

Code available on GitHub

IRODOS | Operating Systems Project

Worked with a team of 5 developers to emulate an operating system shell from scratch.

• Implemented most of the basic functionalities of kernels right from Memory Handling to Process Scheduling with the help of multi-threading libraries.
• Built our own version of a compiler for our OS's unique programming language.

My main role involved writing a robin-round scheduler for handling processes and handling process-threads.

TH 2018

An android app that unites a fun seeking-game with technology to take it to the next level.

• Built a secure android app compatible with versions > Android Kitkat(19.0+)
• Clean minimalist design with easy to use nav-drawer for smooth UX.
• Integrated ZXing QR Scanner Library - to unlock next level
• The app also featured a map of the playing area and list of clues that could only be unlocked progressively.

Code available on GitHub