Design lofi wireframe for data disco workflow

[01painadam]

Goal

Wireframe out the dataset discovery workflow.

Scope

• Wireframe only to illustrate key interactions between the user, UI, and LLM.
• Include dataset search, map visualisation of layers
• Include examples of user interrogation of datasets ("Help me understand this dataset", "how can I use it to do X?")
• Include examples of the LLM raising caveats and cautions that warn the user about limitations of the data
• Include downloading data as an end step for the user (old example here)

User Stories & Acceptance Criteria

• prompt based suggestions When I need a dataset to complete a task but don't know what's available, I want to tell Zeno about my needs e.g. "Monitor the progress of my restoration project", so that I can see a small list of relevant datasets to choose from
• Location based When I am navigating around the map, I want Zeno to dynamically suggest location-specific datasets, so that I can use local datasets in my task
• Summary When deciding what dataset is the most relevant to my needs, I want to see a descriptive summary of the dataset, including critical information like temporal/spatial resolution, coverage, citation text, links, so that I can quickly filter out datasets that do not meet my needs
• handling options When I am looking for relevant datasets and there are multiple candidates, I want to see a list of the most relevant candidates, with a generated explanation of why it is relevant, so that I can make an informed decision and select the one best suited to my needs.
• graceful failure When no datasets are sufficiently relevant to my needs, based on my prompt, I want Zeno to prompt for clarification or refinement of my my initial prompt, so that I can add further details that will help the model identify relevant datasets
• map verification When I am choosing a dataset, I want to add it to the map, to spot check the resolution, coverage, and units so that I can verify that it meets my needs visually
• cautions When I am choosing a dataset, I want to know any critical information about the dataset that I should be aware of, so that I can either choose another, more suitable, dataset or change my methodology to account for nuance in the data
• • download When I have identified a dataset, I want to download zonal stats against a country or raw raster tiles, so that I can run my own analysis off platform

Risks & Questions

• Should cautions be a standard prompt that Zeno suggests? They would function similar to an info button ("!" instead of "i") but the model would tell you about the gotchas and maybe trigger a dialogue about the user's intended use of the data so that it could tell you specifically what to beware of and why
• Similar for links, methodology, tech specs... could these all be standard UI elements that are surfaced whenever a dataset is "in focus"

Solution Hints

• Should we have the model do something like "because you mentioned... X" when justifying a selection, highlighting to the user what the incorrect assumption is in the case of a false postive

Future work