John Nespeca

Designed and implemented agentic AI workflows using LangChain, LangGraph, AutoGen, and MCP for multi-agent orchestration in production conversational and task-driven systems. Led Voice AI development, integrating STT and TTS pipelines to power enterprise voice assistants and improve call automation while reducing human workload. Built enterprise RAG pipelines with embeddings, semantic search, chunking strategies, and prompt engineering for better relevance and factual grounding. Developed and deployed LLM solutions on AWS Bedrock with low-latency, secure integrations to backend services. Implemented Microsoft Dynamics Contact Center solutions using Copilot Studio and Azure AI for omnichannel support with real-time AI assistance. Led platform architecture using Spring Boot and ElasticSearch, mentored engineers, and set prompt-engineering standards across major model providers.

Built LLM-powered conversational and analytical agents for sales intelligence, generating insights from large volumes of voice and text data. Developed Voice AI pipelines using Spark and Vertex AI for speech-to-text transcription and NLP processing to reduce manual review and improve call analysis and summarization. Implemented search and RAG systems using vector databases and embedding-based retrieval for stronger contextual accuracy. Designed agentic AI workflows using LangChain to coordinate multiple AI components for task execution and decision-making. Led LLMOps/MLOps, including model monitoring and CI/CD on AWS and GCP to improve reliability and accelerate deployment cycles.

Developed NLP and conversational AI systems for financial services using Transformers (PyTorch) and LlamaIndex to automate customer interactions and decision workflows. Built search-driven AI solutions using embeddings and semantic similarity with LlamaIndex and PostgreSQL for underwriting and risk assessment. Implemented AI workflow pipelines integrating ML models with Django backend services and CI/CD for reliable, scalable inference. Collaborated with data science and engineering teams to productionize models using MLOps practices and automated deployment pipelines to improve reproducibility and speed up releases.