Harsh Sharma

Harsh Sharma

NLP | Deep Learning Enthusiast | Data Science

SRM Institute of Science and Technology

Biography

I am currently working as BA3 Grad Analyst at Barclays - Pune, India. I have graduated as a B.Tech CSE UnderGrad Student at SRM Institute of Science and Technology, Chennai. Being a Data Science Enthusiast, passionate deep learning developer and researcher, who loves to work on projects belonging to NLP and Data Science Domain. ALways up for hackathons and part of student-led tech clubs, handling executive board positions.

I am also quite active in research domain - especially NLP and Deep Learning.My contribution as a co-author in one of the research papers also got selected for IEEE-INCET 2021 and Springer - Education and Information and technologies (Impact factor 3.6). Currently, I work mentoring juniors in hackathons, Hosting and organizing webinar sessions on deep learning, writing blogs based on deep learning-based natural language processing content, and collaborating with fellow deep learning developers on state-of-the-art projects and research titles constantly.

Recently I interned as MITACS Globalink Research Intern 2022 at University of British Columbia - Okanagan, Canada. I worked under Dr.Fatemeh H Fard for a software engineering based NLP Research Project - Neural Source code intelligence - I developed language models by pre training roBERTa with java bytecode and finetuning it to downstream tasks like Code Summarization and Code Search.

Download my resumé.

Interests
  • Natural Language Processing
  • Artificial Intelligence
  • Deep Learning
  • Time Series Analysis
  • Data Science
Education

  • B.Tech in Computer Science Engineering, 2019

    SRM Institute of Science and Technology, Chennai

  • CBSE - Higher Secondary Education, 2017

    Mahaveer Public School, Jaipur

  • CBSE - Secondary Education, 2015

    C.M. Academy , Ankleshwar

Skills

Deep Learning

Natural Language Processing (NLP) | Time-Series Analysis | Sequence Learning | Sequence to Sequence Networks | Classical ML

Frameworks

Tensorflow - Keras | Pytorch | BERT | Simple Transformers (Hugging Face) | Sentence Transformers (SBERT) | NLTK | Scikit - Learn

Databases

MongoDB (PyMongo) | Firebase (Pyrebase) | MYSQL (SQLite3)

Web Frameworks

Flask | Streamlit | FastAPI | Postman | HTML | CSS |

Programming Languages

Python | C/C++ | R

Cloud Platforms

AWS Basics (EC2-Instance) | Heroku | Azure Basics

Experience

 
 
 
 
 
BA3 Grad Analyst
Barclays
Jul 2023 – Present Pune, India
  • Working in core ETL Team and Devops Team.
  • Working with Innovationshub for Generative AI based projects for production and testing efforts reduction.
 
 
 
 
 
Junior Data Scientist
Kalido
Dec 2022 – Jul 2023 London (Remote from India)
  • NLP - Using Language Models to integrate and enhance existing models for Skill Matching and CV Parser.
  • Working on Building Recommendation Systems using Graph Neural Networks and Language Models.
 
 
 
 
 
MITACS GLobalink Research Intern - SDE - NLP
The University of British Columbia
Jul 2022 – Aug 2022 British Columbia, Canada
  • Got Selected under prestigious MITACS Globalink Research Internship in 2022.
  • Worked as SDE-NLP Intern under Dr Fatemeh H Fard for NLP based project on Neural Source Code Intelligence.
  • Pre-trained roBERTa model for java bytecode to it’s respective documentations using Microsoft’s CodeSearchNET dataset
  • Fine-Tuning it to Downstream tasks like Code summarization and code search.
 
 
 
 
 
Deep Learning Intern - NLP
NeuralSpace
Jul 2021 – Present London HQ, Bengaluru Team
  • Worked on BERT based NER Performance Test on Indic Languages
  • Fine-Tuning Neural Machine Translation on Indic Languages using Marian MT and Pytorch.
  • Currently Working on Exploration of Topic Modelling using Sentence Transformers and BERT.
 
 
 
 
 
NLP Research Intern
National Institute of Technology - MNIT
May 2021 – Present Jaipur, Rajasthan
  • Worked on Multimodal Deep Learning based Neural Network Development on Research Paper
  • Multimodal deep learning - NLP + Image Feature Extraction for Fake or suspicious News Detection
 
 
 
 
 
MLH Pre-Fellowship
Major League Hacking
Jul 2021 – Present New York
  • Pre-Fellow at prestigious Major League Hacking with Summer'21 Batch.
  • Worked On Open Source based Python Package TF-Watcher - A Developer tool used track training of ML models in real time.
  • Bagged the title of “Best Project of the Pod” in tenure of Fellowship.
 
 
 
 
 
Microsoft Learn Student Ambassador - Beta Level
Microsoft Learn Student Ambassador
Jul 2021 – Present New York
  • Beta Level Member at Prestigious Microsoft Learn Student Ambassador
  • Selected as Alpha Level Member
  • Conducted Virtual Workshop on - Data Augmentation in NLP.
 
 
 
 
 
Machine Learning Engineer
Heiphen
Apr 2021 – Present Chennai
  • Part of Start-Up Product incubated by SRM.
  • Worked on Deep Learning Algorithms to derive insights about Fake reviews.
  • Developed Models For Detecting Fake Review and Sentiment Analysis and customized metrics.
  • Deployed and Hosted Via FastAPI on AWS EC2 Instance
  • Working on Custom Named Entity Recognition with Aautomated Resume Parser using BERT.
  • Heiphen is a revolutionary Employee Review Platform created to help employers get genuine employee reviews. We use Machine Learning and comprehensive database-building for thousands of candidate profiles to rate workers according to their proficiency through authentic reviews by their past employers.
 
 
 
 
 
Associate Technical Director and Deep Learning Researcher
Data Science Community SRM
Jul 2020 – Apr 2021 Chennai
  • Started as a Deep Learning Developer where I worked on some state of the art projects.
  • Later worked as Associate Technical Director where I directly managed and organised the events in the community. mentoring juniors and hosting workshops based on data science and deep learning domain.
  • My Contributions were :
  1. Conjexure-Stock Market Forecasting Project ( Conjexure is a machine learning web app for forecasting the stock prices of certain companies into the future, built using deep learning and deployed on streamlit.)
  2. Super-color Project (An ML-based web-application to colorize black and white images, two models trained on TensorFlow and deployed on Streamlit, recognized by the Streamlit developers and included in the worldwide weekly roundup of their Community Forum)
  3. Random Forest: The Optimal Choice For Developers Blog (blog published in the DataX Journal on medium)
  4. Developing Chatbots with RASA-From Intuition to Implementation and Deployment Blog (blog published in the DataX Journal on medium)
  5. Understanding Encoders and Decoders Blog (blog published in the DataX Journal on medium)
  6. Resourceify Contributor (Data Science, Machine Learning, and Artificial Intelligence resources repository on Data Science Community SRM GitHub)
  7. Version Control with Git and GitHub Webinar (Hosted an Internal Webinar session with 100+ attendees)
  8. NeuRes Series (Organized a 6-day long webinar series from 16th to 25th of March on the topic of basics of Deep Learning with over 500+ attendees. Hosted a Session on Deep Learning Practises in Natural Language Processing [NLP]).
  9. Data Science Community Outer Circle Discord (Managed a virtual community of 500+ members via Discord, interacting with deep learning enthusiasts.)
 
 
 
 
 
Community Executive
SRM Machine Intelligence Community
Jul 2020 – Present Chennai
Deep Learning Developer and Researcher at SRM Machine Intelligence Community, An undergraduate research community focusing on the latest happenings in the field of machine intelligence, producing impactful projects, and gaining meaningful insights.

Accomplish­ments & Certifications

An effective deep Learning Pipeline for generation, classification and recommendation in Bloom’s Taxonomy Domains.
Springer - Education and Information Technologies May 2021
Smart and Context-Aware System employing Emotions Recognition, IEEE-INCET 2021.Co-authored a research paper got selected in IEEE-INCET 2021 Research Conference.
IEEE May 2021
See certificate
Coursera
Natural Language Processing in TensorFlow
Coursera Jan 2021
See certificate
National Level 12 th position in Hack the CW Hackathon - Under Top 12 Finalists
HackTheCW Hackathon Dec 2020
See certificate
Secured 13th Position out of 400+ teams in Octhacks 3.0 Hackathon Under Top-20 Finalists
Octahacks 3.0 Hackathon Nov 2020
See certificate
Secured 8th Position out of 150+ Teams in SLAC 2.0 Hackathon, under Top 15 Finalists.
SLAC 2.0 Hackathon Nov 2020
Coursera
Neural Networks and Deep Learning
Coursera Aug 2020
See certificate
Python For Data Science And Machine Learning BootCamp
Udemy Dec 2019
See certificate

Recent Posts

Detailed Comparison of Deep Hybrid Neural Models and Optimizers for Multiclass classification in NLP
Generally, we try to find out the best suitable models, loss functions, and optimizers, which could help to obtain better accuracy by minimum loss. Since we have to deal with textual data, we might also consider it as a sequence problem, with or without contextual relations exploration, which might add a few problems and load while training our model. The main question that remains unanswered — How to select the best model or algorithm for multi-class classification in NLP? Let’s try to explore and answer that.
Harsh Sharma
Mar 30, 2021 1 min read
Understanding Encoders-Decoders with an Attention-based mechanism
When it comes to applying deep learning principles to natural language processing, contextual information weighs in a lot! In the past few years, it has been shown that various improvement in existing neural network architectures concerned with NLP has shown an amazing performance in extracting featured information from textual data and performing various operations for a day to day life. One of the models which we will be discussing in this article is encoder-decoder architecture along with the attention model. The encoder-decoder architecture for recurrent neural networks is actually proving to be powerful for sequence-to-sequence-based prediction problems in the field of natural language processing such as neural machine translation and image caption generation. We will try to discuss the drawbacks of the existing encoder-decoder model and try to develop a small version of the encoder-decoder with an attention model to understand why it signifies so much for modern-day NLP applications!
Harsh Sharma
Feb 1, 2021 1 min read
Developing Chatbots with RASA- From Intuition to Implementation & Deployment
With Advancement in Natural language Processing and various open-source frameworks available, various industries and organizations are trying to implement automated text reply as a response, or, if I have to put in simple words, developing a chatbot. But have you ever wondered, how would it feel like, if you have to make a basic chatbot within 20–30 mins easily! In this article, I would like to cover basic intuitions; what is RASA, About its architecture, and how to implement or develop a chatbot using the RASA framework.
Harsh Sharma
Aug 21, 2020 1 min read
Random Forest - The Optimal Choice For Developers?
Random Forests, the infamous Machine Learning Algorithm, has emerged with a sudden popularity at glance, but, the reason it has become so intriguing to developers these days, is a mystery in itself!
Harsh Sharma
Jul 21, 2020 1 min read

Projects

Recent Projects I worked on

*
All Deep Learning Packages-Tools Research-Related
Personalized-Music-Recommendations-with-Multimodal-Deep-Learning
This repo includes the work of exploration and the usage of multi-modal deep learning – combining or concatenating different features or modes of data – like a song(Audio), its lyrics(Text) associated with its metadata (latent feature vectors) – like a key, tempo, genre, artist information, etc. passed in a combined hybrid neural network-based framework for training and eventually being used to generate personalised recommendations based on this hybrid combination of different modes of data and its associated metadata, rather than using individual non-ensembled networks for the training of different attributes individually. Recommendations being generated from the proposed multi modal deep learning framework provides enhanced personalized recommendations by sequentially processing and combining different modes of data.
Smart Rural Ecosystem
Our Solution Smart Rural Ecosystem aims to dissolve the unwanted interventions of middle-men in their daily livelihood, which depends on day to day heavy tasks they perform and could not get profit they actually deserve. Using appropriate tech stack and dependencies, we would try to propose an ecosystem, which if adopted, could be result in getting achievements and benefit the people living in rural areas.Project repository for course 18CSC206J - Software Engineering and Project Management [SEPM]
Code
TF-Watcher
Python based utility package for tracking Machine Learning Model training progress in Real-Time
Code
TF-Watcher
Computing Semantic Distance between Text Inputs using Siamese Neural Networks
Reimplementation of Paper Siamese recurrent architectures for learning sentence similarity from MIT - https://dl.acm.org/doi/10.5555/3016100.3016291
Code
Conjexure - Stock Market Forecasting
Conjexure is a machine learning web app for forecasting the stock prices of certain companies into the future. Conjexure utilizes the stock prices for Alphabet Inc. (GOOGL) and Apple Inc. (AAPL) for training. This is primarily because the closing stock price graph for these companies over time is quite smooth overall and avoids any sudden ups or downs and follows a healthy upwards trend.
Code
Ed-O-Matic - Smart AI-Digital Based Solutions for Education Analysing
Focused to develop a smart AI Based digital Platform Using Machine Learning /AI and Web based skills to make a secure web application , which aims to analyze the user’s information,it can be a Faculty or Student. Using Appropriate Tech Stack , this app tries to analyse and solve the problem of online assignment upload, to check plagiarism of the assignments sent by the students, the teacher can check if any student had copied the assignments or not through our ML model, to give a short summary and finding the main key points in the assignment so that it could save time for the teacher and student counselling where student can predict a career or predict dropout rate using our ml model..
Code
Emergency Ally
Built using Natural Language Processing and Machine Learning,Emergency Ally is a Comprehensive Web App which can help you provide SOS or Emergency Level Solutions , In Case You encounter or see someone in any Danger or Disaster Situtations.
Code
Emousicolor - Smart and Context-Aware System employing Emotions Recognition
Source Code For Work Described my published research Paper - “Smart and Context-Aware System employing Emotions Recognition”(https://arxiv.org/abs/2106.15101)
Code
Multiclass Hybrid NLP Models Analysis
Detailed Comparison of Hybrid Deep Neural Models and Optimizers for Multiclass classification in NLP. Multi-class classification based on textual data refers to a supervised machine learning task, where we try to predict a certain class for input data, given the input data in raw textual format. Well, thanks to TensorFlow, we have a variety of algorithms available to perform multi-class classification via natural language processing. Here , I have tried to perform a detailed analysis and comparison of various hybrid combination of deep neural networks and choosing different optimizer’s impact on their learning and training of model and hence , their performance and time taken to train.
Code
Multilabel Document Categorization with Semi - Supervised Learning
Devising a semi supervised learning approach for determining topic from corpus.
Code
Rasa COVID-19 Tracker&FAQ Chatbot
A Basic Conversational Chatbot Made with COVID-19 DATA and Integrated with COVID-19 Tracker’s API to get real time details of Cases from all across India.
Code
Sike-Tool - Remote Mental Health Monitoring Tool
Developed using NLP and CV, Our web application aims to analyze the user’s mental health using a common therapy technique–journaling. The user will submit diary entries into our application. In the therapist-patient scenario, they will be required to keep their camera on.
Code

Recent & Upcoming Talks

NLP Workshop in Microsoft Learn Student Ambassadors
Hosted a Workshop/Session explaining on NLP Textual Data Augmentations
Sep 28, 2021 1:00 PM Data Science Community SRM, Chennai
Video Follow
NLP Workshop in Microsoft Learn Student Ambassadors
Talk at Tensorflow User Group Chennai
Gave a guest lecture on Computing Sentence Similarity with Siamese Neural Networks in NLP
May 2, 2021 1:00 PM Chennai
Video Follow
Talk at Tensorflow User Group Chennai
NLP Workshop in Neures
Hosted a Workshop/Session explaining on NLP Hybrid Model Development
Mar 28, 2021 1:00 PM Data Science Community SRM, Chennai
Video Follow
NLP Workshop in Neures

Recent Publications

Quickly discover relevant content by filtering publications.
Harsh Sharma, Stuti Sehgal, Akshat Anand (2021). Smart and Context-Aware System employing Emotions Recognition. In IEEE.

PDF Cite Code DOI

Contact