Goal:

To simulate how UX-optimized content can be structured for LLM-based retrieval systems like RAG (Retrieval-Augmented Generation). The goal was to test chunk segmentation, semantic anchors, and token efficiency — and to understand how a portfolio can become a retrievable, query-aligned source of truth.

Simulation Type:

Self-initiated engineering exercise aimed at replicating core RAG workflows for UX and multilingual content environments.

Context:

As LLMs rely on chunk-based retrieval and semantic proximity, this case explores how a Notion-based personal portfolio can be adapted to support RAG workflows. The structure was tested for standalone meaning, query relevance, and anchor clarity — with UX and multilingual layers in mind.

Language & Token Efficiency Consideration:

This case is provided in English to reflect real-world production scenarios. English tends to generate fewer tokens per sentence compared to Russian or French, making it more efficient in LLM contexts. Token cost and chunk boundaries were taken into account to ensure optimal retrieval performance.

What I Did:

– Chunk segmentation based on semantic units

– Anchor identification and metadata tagging

– UX-aligned phrasing for standalone context

– RAG-suitable structure using Notion blocks

– Semantic clarity testing across query types

– Comparison of language-specific token behavior