Manikanta R

Designed and deployed an LLM-powered underwriting copilot using GPT-4 and LangChain, integrating document processing and enterprise policy retrieval to reduce manual review time by 40% and improve analyst productivity. Built a RAG pipeline using Pinecone vector search, semantic embeddings, BM25 retrieval, and cross-encoder re-ranking across 50K+ documents, improving retrieval quality and reducing hallucination rates by 60%. Developed and integrated FastAPI-based services for real-time fraud detection workflows with 88% accuracy and sub-120ms P95 latency. Implemented MLOps with MLflow, Docker, Kubernetes, and CI/CD for reliable model versioning and zero-downtime deployments. Created human-in-the-loop evaluation workflows with Streamlit and incorporated SHAP explainability into fraud-scoring APIs to reduce false-positive escalations by 25%.

Developed production ML and NLP models using gradient boosting and BERT-based feature extraction, improving predictive accuracy by 20% and achieving AUC of 0.86 for readmission and 0.83 for diabetes progression. Built and deployed real-time ML APIs using FastAPI and AWS SageMaker with sub-200ms response times. Developed demand forecasting using Prophet and LightGBM across 500+ pharmacy locations, achieving 90%+ forecasting accuracy. Applied SHAP and LIME for model evaluation and explainability with collaboration from clinical stakeholders. Automated large-scale data processing and ML workflows using PySpark, Databricks, and Airflow, reducing pipeline execution time by 60% and improving reliability.

Built and maintained Python/Django web applications and REST APIs supporting high-volume operations with over 10,000 transactions per day and 99.5% uptime. Developed an NLP-based intent classification solution to automate support-ticket routing, reducing manual triage workload by 35%. Designed secure REST APIs with JWT/OAuth2 authentication and data validation to reduce data-related issues by 80%. Automated data collection and processing via Python-based scraping pipelines and created analytics/reporting modules using Pandas and NumPy for operational decision-making.