Edward User

Revised with those constraints:

I’m an AI Engineer / Researcher focused on building production-ready AI systems through applied research and industry partnerships at CEIA. Across healthcare, energy, B2B, AI infrastructure, and enterprise consulting, I’ve worked on LLM applications, RAG pipelines, synthetic data workflows, evaluation systems, and multi-agent products.

My work centers on turning ambiguous AI problems into reliable deployed systems. In healthcare, I helped address hallucination and auditability challenges by building a two-agent prior authorization workflow with FAISS-based RAG over regulatory documents, plus an LLM explanation layer over an XGBoost model that improved medical auditor productivity by ~10x. In the energy domain, I developed a vLLM-based synthetic data pipeline and custom evaluation framework for a Portuguese-language domain LLM, contributing to ~5% gains on Portuguese benchmarks and 15–20% improvements on energy-domain evaluations.

I also build agentic and production-facing systems. For a B2B negotiation platform, I used Guidance and logit manipulation to constrain LLM outputs, track conversational state, and improve structured extraction across multi-turn buyer/seller workflows, contributing to a ~30% increase in successful matches and funnel progression. For women’s health, I built a WhatsApp assistant using Google ADK, Qdrant RAG Fusion, FastAPI, and guardrail callbacks.

Earlier, I developed recommendation and RAG components for NiceDay, including a TorchEASE and sentence-transformer recommender deployed with FastAPI and Docker. More recently, during a Meta engagement, I designed and delivered a technical Llama training module for Brazilian enterprise clients and advised teams on on-premise LLM deployment with vLLM, quantization, fine-tuning, and hardware sizing for privacy-sensitive workloads.

My emphasis is reliability, measurement, and systems that work in real production environments.