Apr 2024 – Dec 2024 · 8 mos

Jan 2023 – Apr 2024 · 1 yr 3 mos · Remote

Nov 2021 – Jan 2023 · 1 yr 2 mos · Remote

• At Accern, I built a transformer-based predictive model that achieved a 20%+ increase in performance on two private datasets, and integrated it into the company's inbuilt DAG-based execution platform. I was also responsible for maintaining this internal library, which was used for streaming data and serving models. In addition, I worked on scaling taxonomies using rule generation and named entity recognition, and explored alternative retrieval systems using NER. Overall, my work at Accern focused on building and improving predictive models, as well as maintaining and optimizing internal systems for data management and model deployment.

Nov 2020 – Oct 2021 · 11 mos · Vadodara, Gujarat, India

• Optimal Temperature Prediction for Defect Reduction- Built and deployed 41 models at the
• assembly level to predict the number of temperature-dependent defects on Azure by extractig data from
• DataBricks. Used multiple ExplainableAI solutions that helps stakeholders to identify optimal temper-
• ature therby reducing defects. Optimal temperature lead to fewer defects, thereby saving 500k USD
• ˆ Manpower Requirement Prediction- Used XGBoost to predict manpower requirement based on
• various factors such as planned products, historical manpower requirement, absenteeism, upcoming festi-
• vals/events etc. Built a custom optimizer to suggest optimal manpower based on wage rate and overtime
• cost. Optimal manpower reduced the overtime cost incurred by 134k USD
• ˆ Competitor Analysis Tool- Built a general purpose web scraping tool for competitor analysis in the
• glass industry. Used information from multiple platforms such as linkedin, yahoo finance, etc. to identify
• and extract relevant information based on semantic similarity

Jul 2019 – Oct 2020 · 1 yr 3 mos · Pune

• Click Rate Prediction and Content Optimization- Used Deep Factorization Machines to
• model user click rate and achieved an AP score- 84%. Designed an AI framework which tailor multiple
• fragments together to create a digital asset using Reinforcement Learning - Monte Carlo Tree
• Search approach. Unconstrained uplift of 25% in engagements was consistently achieved across different
• products for the clients from the pharmaceutical industry
• ˆ Content Feature Extractor- Developed a pipeline which leverages ML models to exhaustively tag dig-
• ital fragments with different features such as content category, tone, fragment type, attribute information
• and aesthetic related features such as color, font, font size, etc. MLP was applied on TensorFlow univer-
• sal sentence embeddings and keyword-based features to detect content category and tone with weighted
• F-Score of 80% across multiple brands. CNN was applied to detect type of fragment. Created Named
• Entity Recognition Models using ensemble of Bio-BERT + Sci-Spacy (bionlp13cg, bc5cdr) + med7
• to identify attribute information in a medical text. Performed Color Palette Identification using cluster-
• ing on pixel values. Used clinicalBERT to provide Zero Shot Learning support for Key Message
• identification

Dec 2018 – Mar 2019 · 3 mos · Pune

Feb 2018 – May 2018 · 3 mos · Anand, Gujarat, India

Aug 2017 – Jun 2018 · 10 mos · Anand