MUHEET ALAM

Engineered an end-to-end credit risk decisioning system, utilizing historical lending data to automate and optimize loan approval, review, and rejection processes across diverse loan portfolios.

Developed and validated interpretable machine learning models (logistic regression, tree-based) for default risk prediction, employing cost-sensitive metrics to align with specific business risk appetites.

Segmented model probabilities into Low, Medium, and High-risk categories, resulting in a quantifiable reduction in expected credit losses while simultaneously enabling portfolio growth.

Performed comprehensive threshold and trade-off analyses, clearly demonstrating the impact of risk cut-off adjustments on approval rates, default rates, and overall financial losses.

Designed and deployed a robust serverless Retrieval-Augmented Generation (RAG) API on Google Cloud Run, facilitating on-demand document Q&A with zero infrastructure management and enhanced scalability.

Developed a high-performance FastAPI and LangChain backend to efficiently process business PDFs, generate vector embeddings via OpenAI models, and retrieve context using PineCone vector database for accurate responses.

Containerized the complete RAG pipeline using Docker and implemented automated CI/CD deployment via Terraform and GitHub Actions, streamlining development and operational workflows.

Integrated and configured scalable Google Cloud Platform services including Cloud Storage, Logging, and IAM, ensuring secure model storage, comprehensive monitoring, and robust access control.