Applied AI
Applied AI is part of my current engineering practice: LLM-assisted software delivery, controlled AI access to business systems, retrieval and evaluation workflows, and production use with retained human accountability. The focus is practical implementation and measurable operating value rather than model training or enterprise-wide AI transformation claims.
AI-assisted SDLC / LLM workflows
Current software work uses an AI-assisted SDLC rather than treating an LLM as an isolated coding tool. The workflow is explicit intent and constraints → specification and plan → bounded agent-assisted build → automated tests and machine-verifiable checks → human review and gates → controlled release → production feedback. Codex/LLM assistance is used for implementation, debugging and refactoring; architecture, acceptance, release approval and production accountability remain human responsibilities.
Working principle: Delegate → Review → Own. Delegate bounded engineering work to the agent, require verifiable feedback, review material changes, and own the final design, release and production result.
N.AI — NWH applied-AI reporting
N.AI is an applied-AI reporting assistant integrated with NWH. It analyses user intent, applies safeguarding before retrieval or tool use, and generates responses from controlled business sources through RAG-oriented workflows and controlled API/MCP access. Current production retrieval is API-mediated rather than unrestricted direct model access.
Implementation choices are evaluated against delivery effort and TCO, and responses are checked through both human review and automatic LLM-response testing. Access is currently limited to selected users of the client portal.
Current extension work includes cooperation with suppliers on datasets for SKU vector representations and associated vector values, preparing richer vector-database retrieval for product/SKU questions while keeping source access controlled.