Looking for a job

Vinila Vinila

@vinilavinila

AI/ML Engineer at PNC building fraud scoring for 500K+ daily transactions and production generative AI systems.

At PNC Financial Services, I own a real-time fraud-scoring service for 500K+ transactions a day. Rebuilding it as an XGBoost and LightGBM ensemble cut false-positive alerts by 28% while maintaining the same fraud catch rate.

I also built a RAG assistant over 40,000+ policy and product documents, used by 3,000+ employees. It cut time to an answer from about 12 minutes to under 90 seconds.

At UnitedHealth Group, I worked on clinical, claims, and member-risk models, including an anomaly-detection system screening 2M+ claims a month. I also developed a ClinicalBERT pipeline that cut manual medical-coding review time by 40%.

Earlier, at Citigroup and Honeywell, I worked on retail banking risk and industrial analytics, including predictive-maintenance and equipment fault-detection models. I hold a Master of Science in Computer Science from Florida Atlantic University.

Experience

Work history, roles, and key accomplishments

Current

AI / Machine Learning Engineer

Current

PNC Financial Services

Jun 2025 - Present (1 year 4 months)

Own real-time fraud scoring for 500K+ transactions daily and build agentic LLM applications with RAG, LangGraph, and MCP. Reduced false-positive alerts by 28%, cut time-to-answer from 12 minutes to under 90 seconds, and improved answer faithfulness from 82% to 95%.

Python SQL XGBoost LightGBM Kafka Lang Chain LangGraph MCP Pinecone GPT 4o Ragas Langsmith LoRAs LLaMA 3 VLLM fastAPI Sagemaker MLFlow Kubernetes Evidently AI SHAP LIME SR11 7

Machine Learning Engineer

UnitedHealth Group

Jan 2024 - May 2025 (1 year 4 months)

Developed clinical, claims, and member-risk models on a HIPAA-regulated ML platform. Improved readmission prediction AUC from 0.71 to 0.84, reduced manual coding review time by 40%, and cut call volume by 15% with a member benefits assistant.

Python PySpark Databricks Airflow XGBoost PyTorch ClinicalBERT GPT 4 Azure OpenAI fastAPI Docker Kubernetes GitHub Actions FHIR HL7 PII RBAC

Data Scientist

Citigroup

Jul 2022 - Jul 2023 (1 year)

Improved credit-default prediction accuracy by 14% with ensemble models and SHAP reason codes. Raised confirmed-fraud hits in top 10% of reviewed cases by 22% and cut data latency from 24 hours to 4.

Python SQL Logistic Regression Random Forest SVM SHAP Airflow PySpark PostgreSQL K Means R FM ARIMA Prophet

Data Scientist

Honeywell

Mar 2021 - Jun 2022 (1 year 3 months)

Shortened equipment fault detection from 6 hours to 45 minutes and lowered unplanned downtime by 13% with predictive maintenance models. Developed a CNN defect classifier and forecast spare-parts demand, cutting stockouts by 16%.