Bharath TVS

About

Applied NLP/ML engineer with experience at KeyBank and Novartis building production document intelligence and entity-resolution systems in finance and healthcare. Has delivered end-to-end pipelines (Airflow + AWS) using transformers (DistilBERT/Sentence-BERT), vector search (FAISS/Milvus/Pinecone), and human-in-the-loop labeling to achieve measurable gains (40%+ faster queries; up to 88% F1 and 93% precision/90% recall in entity linking).

Experience

Key Strengths

• Built AI document intelligence pipeline for financial reports/PDFs with entity linking and semantic search across millions of records
• Improved data discoverability and query response times by 40%+ (KeyBank project)
• Fine-tuned DistilBERT with triplet-loss/contrastive objectives for entity disambiguation; achieved ~88% F1 linking entities across documents
• Designed multi-stage entity resolution pipeline combining rule-based matching, fuzzy/TF-IDF similarity, and embedding-based matching; achieved 93% precision and 90% recall
• Production-grade data engineering: modular Airflow/Step Functions pipelines on AWS with data quality (Great Expectations), monitoring (CloudWatch), and CI/CD (GitHub/CodePipeline)
• Strong applied NLP in regulated domains (finance and healthcare/clinical trials), balancing model performance with compute/cost constraints