Resume

Harsh Sharma

Interests: Medical AI, Climate AI

Summary

Artificial intelligence expert with a Master of Science in AI from Boston University and experience researching and developing techniques in natural language processing and Prompt engineering. I am skilled in data science and machine learning with a proven track record of designing and implementing models that improve operational efficiency and reduce costs. I have demonstrated success in applying these skills to real-world problems in vehicle lifecycle management, route optimization, and medical research and seeking a summer intern position where I can use my expertise to advance the field of artificial intelligence and contribute to the organization’s mission.

Resume(pdf)

Checkout my Certifications: link

Education

Boston University

Master of Science in Artificial Intelligence

GPA:3.8/4.0
Expected Dec 2023
Courses: Deep Learning(CS523), Image and Video Computation(CS585), Algorithms for Big Data(CS551)

Indian Institute of Technology, Guwahati

Bachelor of Technology in Engineering Physics

Minor in Electronics and Communication Engineering
Jul 2015 - Jun 2019

Courses: Data Structures and Algorithm, Pattern Recognition and Machine Learning, Parallel Computing

Currently

Graph Prediction on Medical Multimodal data, Advisor: Prof. Vijaya Kolachalama

Alzhimers prediction using Whole slide images as graphs
Creating heterogenous graphs, combining data from multiple sources

Course Project

Multi-object Tracking and Segmentation with BDD100k dataset

Image and Video Computing, CS 585

Increased MOTSP metric by over current SOTA by developing an end-to-end framework combining MaskDINO for segmentation and customized DeepSORT for tracking.

Graphics Generation using Natural Language

Deep Learning, CS523

Implemented a code generation and modification loop using a natural language prompt using Python and Pytorch
Demonstrated adding and removing objects on CARLA simulations using GPT-3

Work Experience

Data Science Intern, WeaveGrid (EV B2B SaaS)

Jun 2023 - Aug 2023

Vehicle Park Time Prediction

Implemented models predicting EV charging behavior, including predicting plug-in times (average error <1.5hrs) and plug-in demand(KWh) (average error <5KWh
Results in reducing peak load by 30% for managed group EV charging, leading to substantial cost savings and significantly improved grid performance.

Machine Learning Engineer, OlaElectric (Electric Two Wheeler Manufacturer)

Feb 2022 - Aug 2022

Vehicle Lifecycle Management

Predicting Battery Faults

Designed Attention LSTM-based models to predict battery faults with a recall rate of 80% using past 100km driving data
The model is comprised of two parts: The first LSTM model focuses on forecasting the time series, the second LSTM focuses on Predicting if the forecasted time series has a fault

Data Pipeline Automation for Efficient Complaint Resolution

Designed and deployed a data pipeline to access all organization data through RESTful APIs, reducing complaint resolution time by ~5 days (40%)
The Django-based application runs ETL jobs scheduled(using Celery) to fetch new data from identified data sources and puts them into the AWS bucket

Real-time Analytics using Apache Spark and Airflow

Designed and implemented a real-time fault monitoring system for electric scooters using Apache Spark and Airflow.
The system provided daily fault statistics and trend analysis to stakeholders via custom email templates
Automated the data collection and presentation process, reducing manual intervention and improving efficiency.

Route Optimization

Implemented a solution that would reduce the cost of transportation for electric bikes by ∼USD 6M per year, a reduction of 30% through the use of mixed integer programming
Resulted in a reduction of per-electric bike transportation cost by 30% and Turned Around Time(TAT) by 20% from placing an order to fulfilling it

Technologies used: Python, Pytorch, Pyspark, Django, PostgreSQL, PowerBI

Machine Learning Engineer, Fractal Analytics (Management Consulting)

Jun 2019 - Feb 2022

Lead Generation for Relationship managers, Standard Chartered (Multinational Bank)

Created ETL pipeline using hive to clean and process transactional data used in 3 downstream applications
Improved write speed into hive tables by 50% by exploiting hive’s properties for the internal module using python
Spot Award: Awarded for going above and beyond for the client

Estimating effect of promotions and Baseline Forecasting, Reckitt Benckiser (Consumer Goods Company)

Created and Deployed an Elasticnet-based regression model that accurately forecasts retail sales and reduces run time by 20 hours; a reduction of 80% compared to earlier deployment
Success of the solution lead to winning contracts for 7 more markets
Star Award: Awarded for out of the box thinking and innovation in deploying solutions

Baseline Forecasting in Covid, Reckitt Benckiser (Consumer Goods Company)

Built a new pipeline with a three-person team to forecast baseline sales for all retail stores in a market near the COVID period
Created novel features based on covid geographic data which helped the model forecast good results even in unstable Covid markets
The created model performed <15% MAPE on unseen near-future forecasts

Entity Mapping, Standard Chartered (Multinational Bank)

Developed entity recognition model using a combination of GloVe embeddings and expectation maximization
Enriched the Embeddings using alternate names extracted from news data and by matching child entities.
Achieved a 90% match score on all client names within the organization

Feedback Capturing, Standard Chartered (Multinational Bank)

Designed and implemented scalable feedback capturing and processing mechanism for Signals generated, which was extended to 16 use cases

Technologies used: Python, Pyspark, Tensorflow, Hadoop, PowerBI

Projects

Contradictory Claims Identification, Coronawhy.org

link Blog

Apr 2020 - Apr 2021

Developed an end-to-end machine learning pipeline to identify contradictory claims from medical research papers
Achieved a ROC-AUC score of 0.8 using a PyTorch and Python-based sentence-encoder model

Research

Bachelor’s Thesis Project, IIT Guwahati, Advisor: Prof. Prabin K Bora

link

May 2018 - Mar 2019

Conducted a study of denoising techniques for speckle noise reduction in OCT images
Designed a novel composite loss function in Keras and Python, resulting in a 3x improvement in the Speckle Suppression Index compared to existing methods

Network Analysis Intern, Center for Development of Advanced Computing