Swapnil Tiwari

Developed and deployed an AI-powered Insurance Policy Insights & Comparison System on Hugging Face using Gradio, implementing RAG with Mistral 7B and FAISS for efficient, semantic comparison and extraction of information from diverse policy documents.

Delivered actionable business insights for a global ride-sharing company by leveraging BigQuery, BigQueryML, and Looker, optimizing operational strategies.

Completed intensive, industry-aligned programs to advance expertise in Machine Learning, Natural Language Processing, and foundational Large Language Models (LLMs).

Currently developing an NLP system for scene-wise dialogue extraction and character development analysis, utilizing sentiment and emotion shifts from theatrical works (e.g., Tamasha).

Spearheaded the design and implementation of a multiclass CatBoost model, integrating advanced embeddings to accurately classify clothing purchase categories, enhancing targeting precision for over 500M+ customers.

Achieved 91% AUC by building a robust model with iterative feature selection, effectively managing concept drift, and integrating an attention layer with embeddings, leading to precise targeting for 500M+ customers across class-wise ventiles.

Generated INR 300K in incremental monthly revenue for 8+ clients by deploying highly accurate model predictions, directly impacting business growth.

Pioneered data-driven digital marketing solutions tailored for FMCG, Real Estate, and Telecom sectors.

Collaborated cross-functionally with clients to develop and execute ML-based audience strategies, optimizing campaigns through A/B testing, which resulted in a significant INR 3M+ increase in monthly net revenue and improved Return on Ad Spend (ROAS).

Led a critical enterprise-wide initiative to combat offer abuse, significantly impacting 10% of 2023 acquisitions and enhancing overall profitability.

Engineered a novel, personalized offer data-driven definition to detect 1.3X more intentional offer exploiters, replacing a 4-year-old legacy system and accelerating identification.

Revitalized predictive gamer models by redefining offer exploitation flags and integrating clickstream signals, achieving an 8% Year-over-Year improvement in AUC.

Drove cross-functional collaboration with marketing, finance, and product teams, managing rapid iterations to implement ROI-driven model cutoffs across 100+ cohorts, which enhanced policy controls and generated $48.2M in savings in 2023.

Developed and deployed a comprehensive MIS dashboard for daily application and gaming impact tracking, resulting in 20+ hours of monthly time savings.

Successfully implemented digital signal analysis to detect offer abuse, leading to $4.7M in direct savings.

Designed and validated 13 behavioral indicators from diverse digital sources (pages, URLs, IPs, devices) to identify gaming tendencies through rigorous hypothesis testing.

Leveraged TF-IDF and word embeddings to identify high-risk pages, then developed a robust Random Forest classifier for detection.

Processed JSON data to generate over 600 variables, employing XGBoost to pinpoint the top 30 critical gaming features.

Constructed a neural network leveraging 43 digital signals to accurately predict digital gaming probability for new acquisitions.

Designed and delivered an end-to-end cloud-based analytics solution for a leading US/UK CPG company, leveraging Google Cloud Platform (GCP) and Looker to enhance operational efficiency.

Developed an ML model on GCP to accurately predict order lead times, providing quantifiable financial insights to inform strategic decision-making.

Automated ML predictions using Airflow DAGs to provide daily estimates of delivery probability, inventory levels, and stockout losses with 50% and 95% confidence intervals, improving supply chain foresight.

Identified and mitigated $5-6 million in 120-day risk-adjusted revenue losses attributed to stockouts.

Created dynamic Looker dashboards to visualize key performance indicators (KPIs), stockout alerts, and financial losses, significantly enhancing business visibility and accelerating data-driven decision-making.

Strategized and executed Data Science solutions impacting 270 million subscribers across diverse business use cases, driving significant value.

Developed and deployed Data Science solutions for critical business areas including prepaid-to-postpaid upselling, churn prediction, and sentiment analysis.

Enhanced model performance by engineering features, optimizing imputation, and fine-tuning, resulting in a 1.4X improvement in churn prediction lift and 1.5X in upselling lift.

Utilized Explainable AI (XAI) to pinpoint key attrition drivers for high-risk customers, enabling the development of targeted and effective retention strategies.