Travoy Spelling

About

ML/NLP engineer focused on production GenAI and data linking systems: built a large-scale RAG pipeline over millions of support docs using LangChain/Pinecone and added a LangGraph-based validation layer to cut hallucinations ~40%. Also built scalable PySpark entity resolution (95%+ accuracy) and fine-tuned Sentence-BERT embeddings with contrastive learning for ~30% relevance lift, with strong CI/CD and observability practices (OpenTelemetry, Prometheus/Grafana).

Key Strengths

• Built production RAG pipeline organizing millions of unstructured documents into a searchable knowledge graph
• Improved information retrieval accuracy by 40%+ and cut support agent response time by ~50%
• Reduced LLM hallucinations by ~40% via context validation layer (confidence scoring + entailment checks)
• Designed high-accuracy entity resolution pipeline (95%+ match accuracy) using PySpark and fuzzy matching
• Scaled large-data similarity pipelines with Spark optimizations (broadcast joins, partition pruning, caching) reducing runtime ~35%
• Fine-tuned Sentence-BERT embeddings with contrastive learning for domain relevance, achieving ~30% lift in retrieval relevance
• Production-grade data workflow engineering with orchestration, containerization, CI/CD, monitoring, and observability

Experience

Education

Skills

Languages