Jeff Cummerata

Architected and led development of a multi-agent AI orchestration platform using CrewAI, AutoGen, and LangGraph, serving 500K+ daily active users. Built advanced RAG pipelines with hybrid retrieval, re-ranking, and query decomposition; implemented agentic workflows with OpenAI Assistants API, Claude tool use, and Gemini function calling with streaming and memory management. Developed fine-tuning pipelines with QLoRA and Unsloth on Llama 3.1 70B and Mistral Large, reducing inference cost by 60%. Engineered real-time voice agents and high-throughput event ingestion/processing using event-driven microservices with Kafka/Kinesis. Led migration to Next.js 15 App Router with streaming SSR and server actions; deployed models on AWS Bedrock and SageMaker with Terraform + ArgoCD. Established observability with OpenTelemetry, LangFuse, and Weights & Biases and mentored a team of 8 engineers.

Built RESTful APIs using Java (Spring Boot) and Python (FastAPI) for a microservice-based web application serving 1M+ monthly users. Used Apache Kafka for streaming and event-driven patterns; designed backend data architecture with MongoDB and GraphQL. Deployed and managed microservices on AWS (EKS, Lambda, S3). Developed AI recommendations services with LangChain (Node.js, later migrated to FastAPI), including multimodal AI features using OpenAI GPT-4, ElevenLabs, Deepgram, D-ID, and HeyGen. Implemented multi-step agent workflows with LlamaIndex and built RAG pipelines with vector search (Pinecone) using multi-route/prompt chains. Added LLM observability and evaluation with LangSmith and wrote unit tests with Mockito/JUnit 5 targeting 85%+ coverage.

Built AI platform infrastructure with Python (FastAPI) and async microservices. Integrated and benchmarked multiple LLMs (GPT-3.5/4, Falcon, Llama 2, Claude 2) using LangChain for task-specific routing. Deployed voice agent bots using Vapi.ai and built multi-step automation workflows with autonomous agent frameworks (BabyAGI, AutoGPT). Developed a cross-platform browser extension (React/TypeScript) and mobile experiences including an Android app (Kotlin) and Apple Watch companion app (Swift/SwiftUI). Integrated Firebase for auth and sync; deployed models via Hugging Face Inference API. Built CI/CD with GitHub Actions and Docker; supported AI/ML deployment using AWS Bedrock and SageMaker.