• Built GenAI pipelines (GPT-4, Llama, Hugging Face) to extract underwriting, claims, and exposure data from complex insurance documents, reducing manual review time by ~70%.

• Designed RAG & Agentic AI systems (Pinecone, Weaviate, LangChain) to ensure LLMs used carrier-approved policy, endorsement, and claims data, cutting hallucinations by 60%+.

• Integrated LLM-generated features into XGBoost & LightGBM models to improve fraud detection, claim severity prediction, and underwriting risk scoring.

• Deployed FastAPI-based AI platforms with PII/PHI redaction, guardrails, and evaluation metrics, enabling secure real-time policy Q&A, clause checks, and document summarization.

Built TensorFlow & PyTorch AI pipelines on DOCSIS, RF, and CMTS telemetry to detect network faults and signal degradation, improving fault-recognition accuracy by 26% across high-density service groups.

• Applied BERT & DistilBERT NLP to modem logs, field tickets, and CMTS alerts, enabling automated root-cause detection and faster NOC triage from unstructured operational data.

• Developed autoencoders, VAEs, LSTM & attention models to predict node health, detect ingress noise, RF leakage, and plant instability before customer impact.

• Deployed real-time AI scoring APIs with SHAP explainability and AWS EMR retraining, enabling proactive, transparent, and scalable network-health monitoring.

• Built Python, PySpark & scikit-learn ML pipelines to score transactions and detect fraud across card, ACH, and digital channels, improving fraud signal throughput by 28%.

• Engineered behavioral, geo-velocity, merchant, and time-window features that significantly improved fraud and credit-risk model accuracy and stability.

• Trained and optimized XGBoost, LightGBM, and ensemble models to reduce false negatives and strengthen high-risk transaction detection under strict risk-governance standards.

• Delivered real-time scoring & model monitoring (Spark Streaming, MLflow, SHAP, drift detection) enabling fast alerts, explainable decisions, and audit-ready fraud operations.

• Built RAG-based semantic search & NLP pipelines to surface insights from public health records, clinical guidelines, and case notes, improving information retrieval speed and accuracy for analysts and care teams.

• Developed ML models for population health, SDOH analysis, risk stratification, and member segmentation, enabling data-driven planning for county healthcare programs.

• Engineered PySpark & Spark SQL pipelines to process large-scale eligibility, encounter, and utilization data, supporting scalable public-sector healthcare analytics.

• Automated reporting and dashboards using Python & Tableau, reducing recurring analytics turnaround time by 30%+ while ensuring responsible and explainable AI.

• Built credit risk & default prediction models using Python, SQL, and scikit-learn on bureau tradelines, utilization, and delinquency data, improving model accuracy by 18% across Equifax’s risk-scoring systems.

• Engineered Spark, PySpark & Hive feature pipelines to create stable, interpretable credit behavior variables for fraud, risk, and creditworthiness modeling.

• Leveraged AWS EMR, S3, Hadoop & Spark to process high-volume bureau data, enabling scalable training, validation, and batch scoring for enterprise credit models.

• Implemented model validation, drift monitoring (PSI/KS), and compliance-ready deployments, ensuring reliable and audit-ready credit decisioning for financial institutions.