Work Experiences
Work Experiences
LinkedIn - Applied Research Engineer, Content Quality (Jul’ 22-Aug’23)
Kindness Reminder
Built an in-house chatGPT like model by fine tuning the GPTNeoXForCausalLM model(with initialized weights from pythia) on LI data to identify if the content to be posted was unkind as per LI’s policy and provide the user with a suggestion on how the content could be transformed to make it kinder.
Review Assist
Performed prompt engineering techniques such as chain of thought to build review assist features using the GPT3.5 API. This saw an increase in the performance of the human labellers and a reduction in average handle time.
Topic Parameterized Misinformation Model
Proposed a novel architecture to include text features from an in-house BERT model and topic distributions from the BART-NLI model to fine tune the in-house BERT model and train a couple of additional layers using a dataset from trending topics in LinkedIn. This is the first such model to be productized on LinkedIn.
EarnIn (June 2024 - Present)
Paycheck Classification:
Working on building a model to classify paychecks from a list of 100k past transactions.This model would use information such as the text descriptions of the transaction, historical similarity between transactions and the amount involved in these transactions to predict if a particular transaction is a paycheck. This classifier would help EarnIn in its recent initiative of ML Payroll that involves migrating the entire process of payroll setup, payroll pattern prediction and early pay cashout to an ML based solution in contrast to the prior heuristic based solution.
Applied Scientist Intern at Amazon (Feb 2022 - July 2022)
Metadata Derived Metadata:
Extract metadata about artists such as descent, age, date of birth, wiki page name, and summary from existing metadata such as artist name and role_type using various NLP and ML techniques. Each of the individually derived metadata had an accuracy greater than 92%.
Enhanced the search/query results in Amazon Music and Alexa, and helped provide information about a song in Amazon Disco, automatic podcast, using the above metadata derived metadata.
Software Development Intern at Microsoft (May 2021 - July 2021)
Repair Condition Prediction in the RNext Tool:
Built an end-to-end automated Machine Learning tool using a model re-trained every 6 hours using NLP for processing the input data and an Lbfgs algorithm on top of the existing model to improve its efficiency.
This model has an estimated improvement in the efficiency of the process by 30%.
PBI Dashboard:
Worked on creating a Power BI dashboard that will help analyze over 100 million requests made across various make services in a month.
This dashboard would give the team actionable insights and help in data-driven decision-making for improving service health and customer satisfaction with a 10x time improvement.
Machine Learning Engineer Intern at Iris News (Feb 2021 - Present)
Built a classification model using a fast incremental gradient method of logistic regression to improve clustering.
The classification model was used after the clustering step to classify the articles in each group to be put into class buckets when used by the user.
This model yielded significant improvements to the overall news article categorization goal with a 100% test accuracy.
Senior Tutor at Avanti an Educational NGO (September 2018 - August 2019)
Student Mentor at Avanti, a student-led NGO that provides access to specialized education focusing on competitive examinations for underprivileged students across India.
Only one among 20 mentors to guide his/her mentee to achieve the prestigious KVPY Scholarship.
Student Mentor at Melvano an EdTech Startup (January 2019 - May 2019)
Member of the Problem Solving Group for competitive exams like IITJEE and Olympiads through Melvano App.
Student Mentor at Unacademy an EdTech Startup (November 2020 - November 2021)
Student Mentor at Unacademy, a leading Edtech Startup for 30+ students preparing for IITJEE, KVPY and Olympiads.