Mohsen Rafiei, Ph.D.’s Post
Data becomes insight when you can see how the pieces connect, but UX still lacks a practical tool for seeing those connections at scale.
I recently worked on a project where the real challenge was not analysis in the usual sense. It was making sense of a huge pile of text. There were many sections, arguments, examples, claims, comments, and pieces of evidence. Each part made sense on its own, but the harder question was this: what is the structure behind all of this?
That is a very familiar problem in UX research too. After interviews, surveys, support tickets, usability notes, app reviews, and old research reports pile up, the issue is rarely that we do not have enough data. The issue is that the data is not structured in a way that can answer higher-level questions.
Raw text can tell you what someone said. Structured text can help you understand how things are connected. That difference matters. Summarizing the material is not enough. I needed to see which ideas supported each other, which claims were connected, where the reasoning was strong, where it was weak, and what larger pattern could explain the whole thing.
This is where approaches like HybridRAG / GraphSearch become useful. Not as magic, and not as a replacement for human thinking, but as a way to map information so we can reason with it.
You can take a pile of text and start turning it into a connected evidence system: concepts, claims, relationships, contradictions, sources, and weights. Once that structure exists, you can ask better questions. Which ideas are connected? Which patterns repeat across different sections? Which claims are supported by multiple pieces of evidence? Where are we seeing the same issue under different names? Where is the gap in the logic?
For UX teams, this is especially useful because many important insights are not sitting in one quote or one document. They are distributed across fragments. One participant describes the behavior. Another explains the emotion behind it. A support ticket shows the consequence. A previous report shows that the issue had appeared before. A product metric shows where it matters.
Of course, the human researcher still has to check the system’s work. Some links are wrong. Some connections are too weak. Some relationships need to be reweighted. Sometimes the system points to a path that looks relevant but is not meaningful for the actual research question.
I put together a free, functional Python prototype for UX researchers who want to experiment with this locally. It runs with LM Studio and local models, has no web UI, no cloud API, and no external database requirement. The basic flow is simple: point it to your UX text files, split them into chunks, extract concepts and relationships, review the suggested links through CSV, and then ask higher-level questions using both semantic retrieval and graph search.
- !graphical user interface, text, application, email