Mounusha Ram Metti

Developed an autonomous AIOps agent for real-time Kubernetes pod anomaly detection and LLM-powered self-healing. Engineered a predictive hybrid ML pipeline utilizing LSTM for time-series forecasting and LightGBM for classification to predict memory outages before they occur. Integrated Google Gemini 1.5 Pro for dynamic remediation and built a purpose-built Go Prometheus exporter for custom metric autoscaling.

Built a full-stack market intelligence platform analyzing 197,430 Swiggy food delivery orders across India. Engineered three proprietary analytical frameworks — a City Expansion Opportunity Index (composite scoring model), a BCG-style Menu Intelligence Matrix, and a Restaurant Health Score viability index — to surface strategic growth signals. Integrated statistical hypothesis testing (Mann-Whitney U, ANOVA), ARIMA forecasting with MAPE validation, RFM segmentation, Pareto analysis, and a live SQL pipeline into an interactive 7-tab Streamlit dashboard with a downloadable 13-sheet Excel KPI report.

Built a production-style AutoML system for tabular data that performs profile-driven preprocessing, tolerance-based model selection, and performance-optimized hyperparameter search, generating fully explainable HTML reports for classification and regression tasks.

Engineered a production-grade multi-agent retrieval-augmented generation (RAG) system for medical query answering with adaptive retrieval thresholds and parallel orchestration. The architecture integrates semantic search, utility-based agent scoring, and controlled LLM generation to deliver grounded, low-latency, and high-precision clinical responses.

Built a decision-optimized customer churn prediction system that converts churn probabilities into profit-maximizing retention actions. The platform uses temporal feature engineering, false-positive cost modeling, and budget-constrained optimization to support realistic production deployment and strategic profit analysis.

Built a real-time object detection inference service optimized for edge deployment, exposing deterministic APIs with temporal activity signals and performance observability. The system converts per-frame detections into structured summaries and short-term context, enabling downstream systems to reason about perception behavior under real-time constraints.

Built a retrieval-augmented generation (RAG) system to answer complex queries using external knowledge sources with grounded, context-aware responses. The pipeline combines semantic retrieval, ranking, and controlled generation to improve factual reliability.