Abhilash Jadhav

Engineered end-to-end ML pipelines on GCP (BigQuery, Vertex AI, Cloud Composer) to predict customer churn with 87% accuracy, reducing attrition by 23% and saving $2.1M annually. Built and deployed a RAG-based LLM chatbot (Llama 3) on AWS SageMaker with vector search (Pinecone), achieving a 92% user satisfaction score and reducing support tickets by 35%. Implemented hierarchical linear modeling and clustering on 2M+ behavioral records to identify retention drivers, improving completion rates by 18%, and deployed MLOps best practices using MLflow and automation to cut deployment time from 2 weeks to 3 days. Developed Tableau monitoring dashboards integrating BigQuery and Snowflake for 15+ KPIs, and led an A/B testing framework for learning interventions across 10K+ students, increasing effectiveness by 27%.

Architected a RAG chatbot on Databricks using fine-tuned LLaMA 3 and FAISS, improving recruiter efficiency by 40% and reducing response time from 4 minutes to 90 seconds. Optimized a global search relevance engine using BERT/TF-IDF and PySpark, boosting recall by 32% and precision by 28% across 15+ markets. Migrated 37 legacy ML projects from dbt to Dataiku with version control and automated testing, cutting pipeline failures by 65% and reducing deployment time by 30%. Deployed a PyTorch recommendation system on GKE serving 2M+ requests/day, increasing CTR by 21% and revenue/user by 15%. Built Kafka-based ingestion and near real-time feature updates for personalization (10TB+ monthly), and implemented CI/CD with Docker/Terraform/Jenkins to achieve 99.7% uptime.

Built an automated fraud detection system using PySpark and XGBoost on 5TB+ weekly transaction data, identifying fraudulent patterns with 94% precision and preventing $8M in potential losses. Designed AWS ML infrastructure using Terraform, SageMaker, and Airflow for 20+ production models, reducing infrastructure costs by 40% via autoscaling and spot instances. Created customer segmentation models (K-means/DBSCAN/PCA) on 15M+ profiles, increasing conversion rates by 26%. Developed interpretable loan risk scoring models using gradient boosting and SHAP for transparent $500M+ lending decisions. Implemented Bayesian time series forecasting in R with 89% accuracy, built NLP pipelines for document classification and sensitive redaction at scale (100K+ documents/month), and established observability/drift monitoring using Prometheus to reduce MTTD by 80%.