Nensi Goswami

Developed an enterprise AI/ML and Generative AI platform leveraging customer usage, billing, and service interaction data to predict churn risk, forecast subscriber and revenue trends, and deliver conversational analytics via Azure OpenAI. Built scalable data pipelines, MLOps workflows, and executive dashboards using Databricks, Spark, MLflow, Power BI, and Azure cloud services. Implemented RAG pipelines using Azure OpenAI, LangChain, and FAISS; developed agentic workflows to automate incident analysis and ticket summarization; processed unstructured documents to support network operations and analytics. Built internal Python APIs to expose AI capabilities to business users and enable interactive front-end applications. Implemented model monitoring and drift detection; evaluated Vertex AI and Azure OpenAI for performance and cost efficiency.

Developed an enterprise Gen AI platform for intelligent document search and conversational knowledge retrieval. Built document ingestion pipelines using AWS Glue and S3; implemented RAG solutions with LangChain, FAISS, OpenAI embeddings; fine-tuned LLaMA-2 and Mistral models using LoRA/PEFT. Deployed secure RAG inference stacks with FastAPI, Lambda, API Gateway, and DynamoDB; designed streaming responses with SSE; implemented multi-stage document chunking and context re-ranking. Built automated evaluation pipelines (RAGAS) and integrated Azure AI Search for secure hybrid search. Designed conversational memory with LangChain and built CI/CD pipelines with GitHub Actions.

Developed a machine learning analytics platform for credit risk, fraud detection, and portfolio monitoring. Built scalable data pipelines using Python, SQL, PySpark, AWS Glue, and Airflow to process transactional data and support predictive analytics. Developed ML models (XGBoost, scikit-learn) for risk and fraud detection; built forecasting models using ARIMA and Prophet. Engineered features, improved data quality, implemented model monitoring, and automated reporting. Delivered explainability with SHAP/LIME and built dashboards in Power BI/Tableau. Supported ETL workflows with Azure Data Factory and MLflow for monitoring and retraining.