Riddhi Pratim Ghatak

• Principal solution architect, leading a team of 4 passionate data scientists, to build Paytm-level Recommendation engine ground-up, which is currently live on all Paytm user base (~240Mn), serving ~150Mn impressions daily to relevant customers, catering to 20+Paytm categories; consistently higher CTR and Transactions than control group is achieved. Currently working towards Real Time serving of the same.
• Solution architect, leading a team of 4 for creating a Multilingual translation product through fine-tuning of LLMs (AI4Bharat, GPT3.5 turbo) to enhance user experience and enable higher reach of on-app communications/offers across various geographies; 35+ BLEU Score achieved w.r.t. Human Translation.
• Ideated the solution process of an Offer creative deduplication tool using OCR, vector embeddings and text matching, driving higher offer creative appeal and freshness and also facilitating offer creators to leverage existing similar banners.
• Lead a team of 2 for building Search text classification using NLP (Fasttext/ BERT/ Sentence2vec) techniques that reduced product recommendation page latency by ~900 Milliseconds

• POC on image-based search capabilities instead of text-based search using Yolo v2 (Object Detection), Color Detection and Image classification.
• Sentiment Analysis on instagram comments using NLP techniques via VADER
• Built Predictive ML Models to enable more accurate targeting of customers during campaign offer roll-out.
• Change agent in organisation to adapt open source ML algorithms in R (XGBoost, Random Forest) instead of traditional modelling techniques like linear and logistic regression in SAS

Designed data science solutions for identifying propensity of incoming calls or work request to result in complaints

• Developed GridSearch algorithm in R for XGBoost hyperparameter tuning to find most optimal model.

• End to end development and implementation of Attrition Prediction Models at global level to proactively identify potential employees in high risk of attrition.
• Devised complete solution for using widely varied compensation related factors cutting across job grades and geography by using benchmarking factors like PPP, inflation, currency exchange rates etc.
• The model reduced the Cost of Attrition and facilitated higher employee engagement.

• Created customer segmentation for identifying different value segments among customers for offer design using K-means clustering.
• Developed Churn Models (Drop in Usage) for accurate customer targeting with attractive offers to mitigate risk of those customer moving out to competition.