Agentic AI

Designed and built agentic portfolio chat on NotNull Online. Before RAG runs, the assistant resolves app commands and navigational intents (login/logout, listing routes, technology-filtered skill views, contact/about shortcuts) and skills-ranking questions against a published content catalog. Follow-up suggestions, entity links, and source citations keep the conversation actionable while the RAG path stays grounded — weak evidence triggers an explicit no-information response instead of invention. The Angular client streams SSE answers, renders markdown with crawlable skill/project links, and supports edit-and-resend turns. Architecture keeps AI providers behind ports so completions and embeddings can swap without rewriting agent routing or retrieval use cases.