Kevin N.

I build AI agents that connect to your business data and tools, RAG systems for reliable knowledge retrieval, and the data pipelines behind them. My work includes a customer-facing clinical assistant and an internal multi-agent software factory. With 10+ years in software engineering, including six years in data engineering, I can take a project from architecture through implementation, deployment, and monitoring. RECENT WORK • ADIA Chat: Built the data integrations, patient-context processing, MCP tools, agent workflows, and chat API for a product deployed to paying customers. Around 10 physicians used it daily, alongside five assistants. Read-only access, PHI-removed views, restricted tool permissions, and deterministic checks controlled what reached the AI. • Software Factory: Built an internal system used by 20 engineers and six non-engineers, including executives. It turned PRDs into agent-driven planning, implementation, testing, and reviewable GitHub pull requests, with validation gates. The team used it to develop the platform's V2. • AI quality and operations: Wrote system prompts and tool instructions, managed large patient contexts, and ran live A/B tests evaluated on accuracy, physician feedback, cost, and latency. Connected monitoring to Datadog and stored performance metrics. • Data platform optimization: At REalyse, reduced infrastructure spending by $11,000 per month and shortened a training workflow from 14 days to five. HOW I CAN HELP • Build AI agents with tool calling, MCP integrations, and access controls • Implement or improve RAG, document ingestion, retrieval, and context management • Connect LLM applications to APIs, databases, and internal business systems • Add prompt evaluations, regression checks, latency/cost tracking, and observability • Build or repair ETL/ELT pipelines and optimize AWS, Spark, and PostgreSQL workloads TOOLS Python, SQL, TypeScript, LangChain, FastAPI, PostgreSQL, AWS, Spark, Kafka, FalkorDB, Docker, Terraform, and Datadog. Tell me what you need the system to do, the data it can access, and where it currently breaks. I can help define a practical scope and build it through to a working deployment.