Yashwanth B

About Yashwanth B

Machine Learning Engineer with 4+ years of experience developing production-grade ML…

Experience

1. Machine Learning Engineer

   PNC

   Aug 2024 — Present · US

   Designed and optimized Retrieval-Augmented Generation (RAG) pipelines for querying unstructured financial documents (loan applications, regulatory filings, KYC/AML records), improving recall by 22% over a BM25 baseline in compliance pilot evaluations.Built an AWS-based document ingestion and indexing pipeline using S3, Textract, Lambda, and OpenSearch, enabling scalable hybrid (keyword + vector) retrieval across compliance and risk workflows.Implemented hybrid retrieval architecture combining BM25 and dense vectorsearch (cosine similarity via FAISS/OpenSearch), improving recall while maintaining exact-match regulatory traceability.Created document-aware chunking strategies (table-aware parsing, header preservation, cross-reference linking), increasing retrieval precision by 18% versus fixed-length chunking in structured evaluation benchmarks.Integrated cross-encoder re-ranking (BERT-class models) to refine top-k retrieval results, improving answer faithfulness and context precision based on RAGAS metrics and analyst review.Developed entity extraction workflows combining AWS Textract and fine-tuned NER models (Hugging Face Transformers) to extract names, dates, and monetary values, reducing manual compliance verification effort.Orchestrated multi-step LLM pipelines using LangChain, including query routing, context assembly within token limits, structured output validation, and audit logging (retrieved chunks, prompts, model/version metadata) to support governance requirements.Built evaluation-driven regression testing (200+ test cases) measuring recall, groundedness, and hallucination risk; integrated into CI/CD workflow to strengthen model validation prior to deployment.