InspectorRAGet

InspectorRAGet, an introspection platform for LLM-based system evaluation. InspectorRAGet allows the user to analyze aggregate and instance-level performance of RAG systems, text generation models, and chat/tool-calling systems, using both human and algorithmic metrics as well as annotator quality.

InspectorRAGet has been developed as a React web application built with NextJS 14 framework and the Carbon Design System.

🎥 Demo

🏗️ Build & Deploy

To install and run InspectorRAGet follow the steps below:

Installation

npm install

⚠️ node version must be 24.0.0 or higher.

Development server

To start InspectorRAGet in development mode, please run the following command.

npm run dev

Build

To build a static production bundle, please run the following command.

npm run build

Production server

To start InspectorRAGet in production mode, please run the following command.

npm start

Usage

Once you have started InspectorRAGet, the next step is import a json file with the evaluation results in the format expected by the platform. You can do this in two ways:

• Use one of our integration notebooks, showing how to use InspectorRAGet in combination with popular evaluation frameworks.
• Manually convert the evaluation results into the expected format by consulting the documentation of InspectorRAGet's file format.

Use InspectorRAGet through integration notebooks

To make it easier to get started, we have created notebooks showcasing how InspectorRAGet can be used in combination with popular evaluation frameworks. Each notebook demonstrates how to use the corresponding framework to run an evaluation experiment and transform its output to the input format expected by InspectorRAGet for analysis. We provide notebooks demonstrating integrations of InspectorRAGet with the following popular frameworks:

Framework	Description	Integration Notebook
Language Model Evaluation Harness	Popular evaluation framework used to evaluate language models on different tasks	LM_Eval_Demonstration.ipynb
Ragas	Popular evaluation framework specifically designed for the evaluation of RAG systems through LLM-as-a-judge techniques	Ragas_Demonstration.ipynb
HuggingFace	Offers libraries and assets (incl. datasets, models, and metric evaluators) that can be used to both create and evaluate RAG systems	HuggingFace_Demonstration.ipynb

Use InspectorRAGet by manually creating input file

If you want to use your own code/framework, not covered by the integration notebooks above, to run the evaluation, you can manually transform the evaluation results to the input format expected by InspectorRAGet, described below. Examples of input files in the expected format can be found in the data folder.

The experiment results json file expected by InspectorRAGet can be broadly split into six sections along their functional boundaries. The first section captures general details about the experiment in name, description and timestamp fields. The second and third sections describe the sets of models and metrics used in the experiment via the models and metrics fields, respectively. The last three sections cover the dataset and the outcome of the evaluation experiment in the form of documents, tasks and results fields.

1. Metadata

{
    "name": "Sample experiment name",
    "description": "Sample example description",
    ...

2. Models

    "models": [
      {
        "model_id": "model_1",
        "name": "Model 1",
        "owner": "Model 1 owner",
      },
      {
        "model_id": "model_2",
        "name": "Model 2",
        "owner": "Model 2 owner",
      }
    ],

Notes:

1. Each model must have a unique model_id and name.

3. Metrics

      "numerical": [
            {
            "name": "metric_a",
            "display_name": "Metric A",
            "description": "Metric A description",
            "author": "algorithm | human",
            "type": "numerical",
            "aggregator": "average",
            "range": [0, 1, 0.1]
            },
            {
            "name": "metric_b",
            "display_name": "Metric B",
            "description": "Metric B description",
            "author": "algorithm | human",
            "type": "categorical",
            "aggregator": "majority | average",
            "values": [
                  {
                        "value": "value_a",
                        "display_value": "A",
                        "numeric_value": 1
                  },
                  {
                        "value": "value_b",
                        "display_value": "B",
                        "numeric_value": 0
                  }
                ]
            },
            {
            "name": "metric_c",
            "display_name": "Metric C",
            "description": "Metric C description",
            "author": "algorithm | human",
            "type": "text"
            }
      ],

Notes:

1. Each metric must have a unique name.
2. Metric can be of numerical, categorical, or text type.
3. Numerical type metrics must specify range field in [start, end, bin_size] format.
4. Categorical type metrics must specify a values field. Every entry must have both a value (the string label) and a numeric_value (an integer or float). The numeric_value encodes the ordering and distance between categories — it is used by aggregation (mean, median, majority), inter-annotator agreement distance, chart axis ordering, and filter range sorting. Without it, all of those computations fall back to parseFloat(label), which returns NaN for non-numeric strings and silently corrupts statistics and visualisations. Assign values so that higher = better (e.g. poor=0, acceptable=1, good=2). The platform sorts values ascending by numeric_value, treats the highest as maxValue, and uses it as the top of all normalisation and ranking scales.
5. Text type metrics are only accessible in the instance-level view and are not used in any aggregate statistics or visual elements.

4. Documents

      "documents": [
            {
                  "document_id": "GUID 1",
                  "text": "document text 1",
                  "title": "document title 1"
            },
            {
                  "document_id": "GUID 2",
                  "text": "document text 2",
                  "title": "document title 2"
            },
            {
                  "document_id": "GUID 3",
                  "text": "document text 3",
                  "title": "document title 3"
            }
      ],

Notes:

1. Each document must have a unique document_id field.
2. Each document must have a text field.

5. Tasks

      "filters": ["category"],
      "tasks": [
            {
                  "task_id": "task_1",
                  "task_type": "qa",
                  "category": "grounded",
                  "input": [
                        {
                              "speaker": "user",
                              "text": "Sample user query"
                        }
                  ],
                  "contexts": [
                        {
                              "document_id": "GUID 1"
                        }
                  ],
                  "targets": [
                        {
                              "type": "text",
                              "value": "Sample response"
                        }
                  ]
            },
            {
                  "task_id": "task_2",
                  "task_type": "generation",
                  "category": "random",
                  "input": [
                        {
                              "speaker": "user",
                              "text": "Hello"
                        }
                  ],
                  "targets": [
                        {
                              "type": "text",
                              "value": "How can I help you?"
                        }
                  ]
            }
      ],

Notes:

1. Each task must have a unique task_id.
2. Task type can be qa (single-turn retrieval QA), generation (text/JSON generation), rag (multi-turn retrieval conversation), or tool_calling (function-calling evaluation).
3. For qa, generation, and rag tasks, input is an array of utterances where each utterance has a speaker (user or agent) and a text field.
4. For tool_calling tasks, input must be an array of messages following the OpenAI chat completion format.
5. For qa and rag tasks, the contexts field is an array of document references (subset of documents) available to the model.
6. targets is an array of expected outputs. Each target is a typed object: { "type": "text", "value": "..." } for text outputs, or { "type": "tool_calls", "calls": [...] } for tool-calling ground truth.
7. category is an optional field for grouping similar tasks.
8. filters is a top-level field (parallel to tasks) specifying an array of task fields to expose as filters during analysis.

6. Results

"results": [
      {
            "task_id": "task_1 | task_2",
            "model_id": "model_1 | model_2",
            "output": {
                  "type": "text",
                  "value": "Model response text"
            },
            "scores": {
                  "metric_a": {
                        "system": {
                              "value": 0.233766233766233
                        }
                  },
                  "metric_b": {
                        "system": {
                              "value": "value_a | value_b"
                        }
                  },
                  "metric_c": {
                        "system": {
                              "value": "text"
                        }
                  }
            }
      }
]

Notes:

1. results must contain one entry for every model defined in models and every task in tasks. Total number of results equals number of models (M) × number of tasks (T).
2. Each result must be associated with a single task and a single model.
3. output is a typed object representing the model's response. For text responses use { "type": "text", "value": "..." }. For tool-calling tasks use { "type": "tool_calls", "calls": [...] }.
4. scores captures ratings for the model on a given task, for every metric specified in the metrics field.
5. Each metric score is a dictionary with evaluator/worker IDs as keys. In the example above, system is the worker ID for an automated scorer.
6. Each per-worker score must be a dictionary containing at minimum a value key with the numeric or categorical rating.

Citation

If you use InspectorRAGet in your research, please cite our paper:

@misc{fadnis2024inspectorraget,
      title={InspectorRAGet: An Introspection Platform for RAG Evaluation},
      author={Kshitij Fadnis and Siva Sankalp Patel and Odellia Boni and Yannis Katsis and Sara Rosenthal and Benjamin Sznajder and Marina Danilevsky},
      year={2024},
      eprint={2404.17347},
      archivePrefix={arXiv},
      primaryClass={cs.SE}
}