Meghana Reddy

Hyderabad

• Own the product-recommendation system (PyTorch two-tower model) serving 9M users; lifted click-through 22% and add-to-cart 11% over the prior heuristics.
• Built a real-time inference service on AWS SageMaker + Triton holding p99 latency under 80ms at 4,000 requests/second during peak sale traffic.
• Designed the Airflow + Spark feature pipeline feeding a Feast feature store, cutting feature-freshness lag from 24 hours to 15 minutes.
• Shipped an LLM-powered support-triage classifier (fine-tuned transformer) that auto-routes 68% of tickets, saving the CX team ~₹40L/year in handling cost.
• Standardized MLOps with MLflow and Kubeflow pipelines, cutting model-to-production time from 3 weeks to 4 days and mentoring 2 junior engineers on it.

Bengaluru

• Built a fraud-detection model (XGBoost + scikit-learn) on 12M transactions that cut fraud losses 31% while holding false positives under 0.4%.
• Deployed batch and streaming scoring on GCP Vertex AI with automated drift monitoring, catching 3 silent data-quality regressions before they hit revenue.
• Reduced training cost ~₹5.5L/year by moving feature computation to BigQuery and right-sizing GPU jobs.