Ronak Doshi

Python-Based EEG and Deep-Learning worklfow

It is a prototype web-based application allowing drag-and-drop creating, editing, and running workflows from a predefined library of methods. Workflows are designed using individual component blocks that have completely configurable inputs, outputs and properties. The Blocks can be combined and rearranged and executed.

Description:

• Refactored the system from Java to Python
• Created deep learning blocks for EEG processing and classification using Tensorflow
• Created API endpoints for the system using Flask
• Applied concepts of Graph theory for the workflow execution

Check out:

• Certificate
• Github Repository
• GSoC Project Abstract

Selection: Solving the given problem statement followed by 3 rounds of interview (2 Technical, 1 HR)
Description:

• In chargeof the backend for one of their leading product called MIT
• MIT is a computer vision tool for ranking global fashion trends
• Restructured, Redesigned and Coded the backend for MIT from Scratch
• Created31 API endpoints using Flask(python)
• Shifted from server-side rendering to client-side rendering

Check out:

• stylumia.com
• Internship Certificate

In the present day social media provides opportunity to market products via influencers, owing to their massive popularity, provide a huge potential customer base generating higher returns of investment in a very short period. However, it is not straightforward to decide which influencers should be selected for an advertizing campaign that can generate maximum returns with minimum investment.

• In this work, we built an agent-based model (ABM) that can simulate the dynamics of influencer advertizing campaigns in a variety of scenarios and can help to discover the best influencer marketing strategy.
• Designed a probabilistic graph-based model that incorporates real-world factors such as customers' interest in a product, customer behavior, the willingness to pay, a brand's investment cap, influencers' engagement with influence diffusion, and the nature of the product being advertized viz. luxury and non-luxury.
• Using customer acquisition cost and conversion ratio as a unit economic, we evaluate the performance of different kinds of influencers under a variety of circumstances. Simulations were performed on real world social graphs of Twitter and Google plus containing 80k nodes, 1.7 million edges.
• Our simulations reveal the importance of different influencers (e.g. micro-influencers and celebrities) in varying circumstances of advertizing.

Check Out:

• Paper: arxiv.org/abs/2106.01750

Tradezi is a virtual trading platform developed as a learning tool to make the process of entering the investment cycle a little easier by letting users invest risk free in real stocks using virtual money.

• It is an online learning platform that provides real-time stock market experience for beginners to explicate the insides of stock market and its role in economic growth.
• Platform supports virtual trading, portfolio management and historical stock data analysis.
• Deployed following DevOps pipeline of building, testing, continuous integration using Jenkins, continous deployment using Ansible, and monitoring using ELK stack.

Check Out:

• Website: tradezi.co.in
• github.com/Tradezi

EEG signals play an essential role in detecting onsets of seizures, head injury, stroke, brain tumors, and other conditions such as dizziness, headache, dementia, and sleeping problems. To extract relevant features from the EEG signal, research in deep learning and artificial intelligence have come up with algorithms that help determine such diagnosis efficiently and more accurately. Hence we propose an online drag-and-drop visual tool to design and execute such deep learning pipelines. Additionally implemented a module consisting of blocks specifically for EEG signal processing and classification.

• Here you can build workflows using individual component blocks that have completely configurable inputs, outputs, and properties so that the blocks can be easily combined and rearranged at runtime without modifying the code.
• This will help researchers quickly and effortlessly build deep-learning architectures to extract valuable information from EEG signals
• Tested the system on the P300 EEG waveform and were able to procure promising results

Check Out:

• Website: eeg-workbench.herokuapp.com
• GSoC Project Page
• github.com/ronak66/EEG-Workflow-System

An intuitive writing software with note management facilities. All the notes are stored in our servers. They can be accessed anytime from anywhere ! It is a platform where people can share, and access the classroom notes. It has a simplistic interface which is easily understandable. The aim of the project is to facilitate the process of writing notes in classes. Professors, teachers and students are hugely benefited by this as it saves a lot of time.

• Real-time saving of in-class notes written by professor which enables students to focus on the teaching rather than writing note
• Beta stage has been deployed and is being used by faculty of IIIT Bangalore

Check Out:

• Website: webboard-old.herokuapp.com

Seizure episodes are frequently observed for adults and children suffering from medically refractory epilepsy. Inorder to acquire accurate localization of seizure onset Ictal prefusion study is performed where a radioactive Tc-HMPAO is injected before seizure onset. However, the onset of a seizure is a highly unpredictable event and makes it difficult to administer the tracer manually within the ideal time frame.

Hence a complete autonomous injection of radioactive tracer element without manual intervention is expected to offer a highly accurate epileptical focus region and aids in further management of the patient. The proposed injection system works on the seizure prediction model from the EEG signals to release the dosage, making the system completely autonomous in action. The syringe based injection system was characterized to emulate dosage release action with minimum volumetric error, and low injection time, on predicting seizure Ictal event from the EEG signal.

• The autonomous injection system was experimneted on medical mannequin ( venipuncture training pad)
• Accepted as paper in 43rd IEEE EMBC 2021 Conference - Acceptance Letter (Yet to be Published)

It is a research project under Prof Dr. Madhav Rao
It is a surgical operating microscope with the primary aim to enhance the range of view for surgeons during surgical practices. This device is built in the form of a continuous series of 3D printed hollow structures that rest on top of each other providing a hollow space. This hollow structural configuration is needed in order to insert other surgical tools to reach and operate a specific region without the need to disturb the probe.

• Medical student from NIMHANS conducted 15 real Human Brain Dissections using our surgical snake
• Published as paper in 17th IEEE ICMA 2020 Conference, held at Beijing,China - IEEE Explore
• Project got featured in one of the most pretigious newspaper in India called The Times Of India newspaper. Article link: TOI article

• Given a image of a segmented cell from the thin blood smear, detects whether it is infected by malaria parasite or not, using Machine Learning and Image Processing
• Model is able to predict correct outcome with accuracy score of 95.87%

Check out:

• Project Report
• github.com/ronak66/Malaria-Parasite-Detection
  

It is a video analytics solution for a retail shop which has a camera placed at the entrance. Given a snap i.e. a image frame sampled every 5 seconds assuming people enter one-by-one at the entrance, it finds if there is a person in every snap, and if so what are they wearing - “Formal Shirt/ T-Shirt/ Saree/ Kurti / None of the above”.

Check out:

• Project Report
• github.com/ronak66/Dress-Detection

• A music recommendation system that recommends songs according your current mood. Given a current song being played, it learns and recommends songs that you would like to listen to next
• Achieved by breaking the available song data-set into K clusters and recommend song based on popularity and it’s distance from current song. Applied reinforcement learning technique for better convergence

Check out:

• Project Report
• github.com/ronak66/Song-Recommendation-System