Benjamin Dong

Built and scaled production-grade AI systems centered on LLMs, RAG, and multi-agent orchestration to automate enterprise workflows and decision-making. Designed LangGraph/LangChain agent frameworks and RAG pipelines over 500K+ documents, significantly reducing manual review while achieving high retrieval precision through evaluation-driven optimization. Engineered high-throughput FastAPI microservices on AWS ECS with low-latency performance, robust CI/CD, and enterprise-grade reliability. Led full-stack delivery of real-time AI dashboards and partnered cross-functionally to ensure security, compliance, and GDPR-aligned deployments across large-scale production environments.

Designed and launched a GenAI-powered movie discovery and recommendation platform leveraging LangChain, LLM embeddings, Pinecone vector search, and BERT-based intent models to deliver highly personalized content discovery. Built RAG-style pipelines and multi-step agent workflows for mood detection, genre classification, and similarity scoring, significantly improving recommendation relevance and reducing noise. Engineered low-latency async FastAPI services on AWS EKS and delivered conversational, streaming UI experiences with React and Next.js, driving measurable gains in engagement, precision, and system reliability.

Designed and deployed a virtual health assistant leveraging LLMs and RAG for real-time symptom analysis and patient triage across 200K+ clinical records, supporting thousands of daily sessions with risk-aware, clinician-validated guidance. Built NLP pipelines with LangChain, BERT, and Hugging Face to improve symptom and intent classification accuracy by 40%+, and integrated high-recall semantic retrieval over medical literature and diagnosis rules with downstream safety filtering. Engineered secure, low-latency FastAPI microservices on AWS ECS Fargate, implemented multi-step agent workflows with clinical rule cross-checks to reduce hallucinations, and delivered HIPAA-aligned deployments through encrypted data handling, audit logging, and controlled access.