Jan 2025 – Present · 1 yr 2 mos · Bengaluru, Karnataka, India

• Brains behind building the last analyst the world needs (ChatGPT on steroids :))

Oct 2023 – Dec 2024 · 1 yr 2 mos · Bengaluru, Karnataka, India

• Utilized NLP techniques including topic modeling, semantic similarity, and semi-supervised classification to develop an automated classification system for Amazon products in the Marketshare project, reducing the onboarding time from 8 weeks to 1-2 weeks thus enhancing efficiency.
• Drove the development of an Automated SKU matching tool leveraging NLP (NER, semantic similarity) and vision techniques, achieved 90%+ accuracy improvement, 40% reduction in matching load, and streamlined quality control across 100s of retailers, cutting manual operations by 80% and reducing turnaround time by 12 weeks.
• Fine-tuned Mistral 7B using LORA to create a universal classifier with Amazon level hierarchy to optimize classification and product categorization for various client facing products (DSA & Marketshare).
• Developed a POC for automating tag and feature identification in GraphDB using techniques including topic modeling, prompt engineering, and LLM fine-tuning. This initiative aimed to eliminate manual effort to create tags thus enhancing efficiency and accuracy in tagging .
• Utilized classical ML techniques such as linear regression and PCA to develop an automated Sales RCA model to explain the factors responsible for the shift in sales to generate actionable items for our clients.

Apr 2022 – Sep 2023 · 1 yr 5 mos · Bengaluru, Karnataka, India

Dec 2021 – Feb 2025 · 3 yrs 2 mos

Jul 2021 – Apr 2022 · 9 mos · Hyderabad, Telangana, India

• Developed price elasticity API for AJIO by modeling competitive and festive effects and improved the sales to inventory ratio by 15%
• Improved the demand forecasting accuracy by 3% using the novel approach of hierarchical transfer learning using BiLSTM
• Modeled shap value-based linearization of black-box models and achieved high accuracy with high explainability (Conference paper)

May 2020 – Jul 2020 · 2 mos · India

• I worked on the novel problem of causality detection and predicting the modified demand. After trying linear methods I reached to the conclusion that it is important to figure out the products which are related. So I used Approximate Nearest Neighbors to perform clustering. I perfomed feature engineering and feature selection using various techniques and then used multivariate regression models for prediction. I used many other non linear methods like LSTM, XGBoost, Random Forest, CNN After performing hyperparameter optimization and several trail and test runs i finalized an ensemble architecture with 2 base learners and 2 meta learners. It was a very great learning experience.

Jul 2020 – Dec 2020 · 5 mos · Bengaluru, Karnataka, India

• Modeled ways to increase conversion in Bangalore. Conducted the data exploration on Geo level, User-level, and Store level. Divided Bangalore into 6 categories of geo and generated hypotheses and plans of action for each category. Predicted the change in conversion using ElasticNet and lasso as base learners and XGBoost as Meta learners. Verified the results by performing A/B testing.

May 2019 – Dec 2019 · 7 mos

• I was a part of a 13-member team responsible for placements of 300 final year students.
• Worked with a team to organize an internship season for 450 pre-final year students.
• Assisted and coordinated with the core team in various aspects including logistics and operations.

May 2019 – Jul 2019 · 2 mos · Kolkata Area, India

• Algorithm designing to convert any randomly arranged notepad datafile to its proper excel datasheet with proper alignment of data in a tabular
• format for further use and analysis.
• Combining multiple dynamically selected data ranges from different excel sheets for data analysis.
• Plotting and analysis of the data of any excel sheet. The plots taken into consideration are bar plots and pie charts.
• CAGR calculator for analyzing the annual growth rate of select parameters.
• GUI development combines all the above-mentioned functions into a single GUI.