Molli Dinesh

About

Built an AI-driven insurance policy summarization platform at Marsh, taking it end-to-end from messy PDF ingestion/OCR and custom extraction through LLM fine-tuning and AWS SageMaker deployment. Delivered measurable impact (25% reduction in manual review time, 99% uptime) and demonstrated strong production MLOps/LLMOps practices with Airflow/Step Functions orchestration, rigorous evaluation (ROUGE + human review), and continuous monitoring for drift, latency, and hallucinations.

Experience

Education

Key Strengths

• Built and deployed end-to-end LLM summarization system (ingestion → OCR/extraction → fine-tuning → SageMaker deployment)
• Improved insurance policy review efficiency by 25% with production system
• Achieved 99% uptime for deployed AI system
• Strong document AI problem-solving (mixed-layout PDFs, low OCR confidence, tables/handwritten notes) with layered preprocessing and adaptive OCR routing
• Practical mitigation of domain drift and hallucinations via domain-specific fine-tuning and expert feedback loops
• Production-grade orchestration of ML/LLM pipelines (Airflow, Step Functions, SageMaker Pipelines) across Spark/Kafka workflows
• Reliability engineering for pipelines using monitoring/alerting (Prometheus/Grafana) and workflow optimization (parallelization, caching, resource tuning)
• Clear evaluation discipline: offline benchmarks + human-labeled evaluation + stress tests + shadow mode + continuous production monitoring