I design and build production-grade machine learning systems and autonomous agentic diagnostics engines. My work focuses on real-world engineering constraints: sub-millisecond inference latencies, class-imbalance resilience, and causal reasoning across complex distributed telemetry streams.

• 🔬 Production ML & Architecture: Building multi-tier defense engines combining deep sequence modeling (CRNN + Attention), in-memory graph topologies, and calibrated ensemble classifiers.
• ⚡ Telemetry & Agentic Systems: Creator of AgenticEV, using the Model Context Protocol (FastMCP) and causal inference graphs to diagnose EV charging failures across OCPP 1.6-J and OCPI protocols.
• 🐳 End-to-End Systems Engineering: Delivering production-ready code with FastAPI, Docker / Docker Compose, Streamlit SOC Dashboards, and automated Pytest suites.

Autonomous AI-Driven Root Cause Analysis (RCA) & Causal Inference Engine for EV Fast-Charging Networks

• Domain: Resolves the 18%–25% charging failure rate across distributed EVSE hardware, vehicle BMS contactors, and OCPI roaming protocols.
• Tech Stack: Python • FastMCP • Playwright • FastAPI • Pytest • Word/Excel Synthesis
• Key Highlight: Uses FastMCP (Model Context Protocol) to empower LLM agents with autonomous telemetry diagnosis tools, temporal anomaly detection, and deterministic fault attribution.

Enterprise Real-Time AI Fraud Prevention System for Digital Payment Networks

• Domain: Real-time fraud detection under extreme class imbalance (<0.2%) and sub-millisecond latency budgets for instant payment rails (UPI / Faster Payments).
• Tech Stack: TensorFlow • CRNN + Bahdanau Attention • In-Memory Graph • FastAPI • Docker • Streamlit
• Key Highlight: Sub-1.2 ms inference latency (~865+ req/s throughput) achieved via compiled @tf.function graph tracing, multi-tier defense fusion, and dynamic payment graph mule-ring detection.

Clinical AI Diagnostic Decision Support System for Breast FNA Cytology

• Domain: Minimizes fatal False Negatives in cytological evaluation of breast fine needle aspirates (FNA).
• Tech Stack: Scikit-Learn • HistGBM + RF + MLP Ensemble • Platt Scaling • FastAPI • Docker • Streamlit
• Key Highlight: Soft-voting clinical ensemble reaching 99.67% ROC-AUC and 95.24% sensitivity, equipped with Z-score biomarker explainability (XAI) and a medical-grade SOC dashboard.