This repository is a demonstration project to show how LLMs can be used to help scattering facility users analyze data. It was created out of the need to analyze a large number of data sets for a real experiment. It is based on the idea that a list of properly named, well defined, and well documented tools can easily be interpreted by an LLM, which can then call them according to a user's input. This version has LLM instructions for both GitHub Copilot and GEMINI. This project is still in the prototype phase. Much remains to be done in terms of flexibility and reporting. Please feel free to contribute your own tools.

New to this repository? Start here:

The easiest way to get started and use this project, follow these steps:

1. Install VS Code
2. In VS Code, enable GitHub Copilot by logging in to your GitHub account
3. Follow the installation steps below
4. Make sure your data is available on the computer you are working on
5. Modify the config.ini file and make sure the combined_data_dir and partial_data_dir point to your data (they can be in the same directory).
6. Make sure the results_dir in the config.ini file is somewhere you like
7. If you are from another facility or you data files have a different naming convention (and if you are brave), you can change the combined_data_template in config.ini to define how your data files are named.

From the command line

# In Python or Jupyter
from analyzer_tools.welcome import welcome
welcome()

# Or from command line
python analyzer_tools/cli.py

This will show you all available tools and help you get started!

• 🔍 Data Quality Assessment: Check partial data consistency before combining
• 📈 Model Fitting: Fit reflectivity data to theoretical models
• 📋 Automated Reporting: Generate comprehensive analysis reports
• 🛠️ Model Management: Create and modify fitting models
• 📈 Result Analysis: Evaluate fit quality and parameter uncertainties

1. Clone the repository:

   git clone <repository-url>
   cd analyzer

2. Set up Python environment:

   python3 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   pip install --upgrade pip
   pip install -r requirements.txt

3. Install the package (optional):

   pip install -e .  # Installs CLI commands: run-fit, assess-partial, create-model

# List all tools
analyzer-tools --list-tools

# Get help for specific tool  
analyzer-tools --help-tool partial

# Show analysis workflows
analyzer-tools --workflows

# Interactive tool selection
python -c "from analyzer_tools.welcome import help_me_choose; help_me_choose()"

The config.ini file is used to configure the paths for the data and results directories.

[paths]
# This is the top-level directory where the fit results will be stored
results_dir = /tmp/fits
# This is the directory containing your final reduced data
combined_data_dir = data/combined
# This is the directory containing the partial data, one file for each run
partial_data_dir = data/partial
# This is the directory where we will write the reports for each data set
reports_dir = reports

You can edit this file to change the default locations for your data and results.

Data locations are configured in config.ini. Default structure:

• Default Location: data/combined/
• Configurable: Set combined_data_dir in config.ini
• Template: Configurable via combined_data_template (default: REFL_{set_id}_combined_data_auto.txt)
• Use: Final reduced data ready for fitting

• Default Location: data/partial/
• Configurable: Set partial_data_dir in config.ini
• Format: REFL_<set_ID>_<part_ID>_<run_ID>_partial.txt
• Use: Individual data parts before combining (for quality assessment)
• Structure: Usually 3 parts per set, all with same set_ID belong together

All data files contain 4 columns: Q, R, dR, dQ

• Q: Momentum transfer (1/Å)
• R: Reflectivity
• dR: Reflectivity uncertainty
• dQ: Q resolution

# Assess overlap quality between data parts
python analyzer_tools/partial_data_assessor.py 218281

# Check the report
open reports/report_218281.md

# 1. Run the fit
run-fit 218281 cu_thf

# 2. Assess results
assess-result results/ 218281 cu_thf

# 3. Check reports directory for results
ls reports/

# 1. Create model variant (note: quote the parameter adjustment)
python -m analyzer_tools.create_temporary_model cu_thf cu_thf_wide --adjust 'Cu thickness 300,1000'

# 2. Test with new parameters
run-fit 218281 cu_thf_wide

# 3. Compare results
assess-result results/ 218281 cu_thf_wide

Edit config.ini to customize paths and settings:

[paths]
# Fit results storage
results_dir = /tmp/fits                    

# Data directories (customize for your setup)
combined_data_dir = data/combined          # Final reduced data
partial_data_dir = data/partial           # Individual data parts  

# Output directories
reports_dir = reports                     # Analysis reports

# File naming template for combined data
combined_data_template = REFL_{set_id}_combined_data_auto.txt

🔧 Customizing Data Locations:

• Set combined_data_dir to point to your reduced data location
• Set partial_data_dir to point to your partial data location
• Modify combined_data_template if your files use different naming
• All tools automatically use these configured paths

Models are Python files in the models/ directory. Each must contain a create_fit_experiment function returning a refl1d.experiment.Experiment object.

Available models:

• cu_thf: Copper thin film model

Create new models:

# Copy existing model as template
cp models/cu_thf.py models/my_model.py

** Create a model file that can be loaded in the refl1d interface **
python analyzer_tools/create_model_script.py my_model data.txt

This repository is designed to work seamlessly with AI assistants. The tool registry system allows AI assistants to:

• Automatically discover available tools
• Understand tool capabilities and usage
• Guide users through appropriate workflows
• Provide contextual help and examples

For AI assistants: Import the registry to access tool information:

from analyzer_tools.registry import get_all_tools, get_workflows, print_tool_overview

This package includes an MCP (Model Context Protocol) server that allows Claude Desktop to directly call the analysis tools. To set this up:

1. Install the package and dependencies:

cd /path/to/analyzer
pip install -e .
pip install fastmcp

2. Locate your Claude Desktop configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

3. Add the analyzer-tools server configuration:

Edit the config file and add (or merge with existing) the following:

{
  "mcpServers": {
    "analyzer-tools": {
      "command": "fastmcp",
      "args": ["run", "analyzer_tools/mcp_server.py"],
      "cwd": "/path/to/analyzer",
      "env": {
        "PYTHONPATH": "/path/to/analyzer"
      }
    }
  }
}

⚠️ Important: Replace /path/to/analyzer with the actual absolute path to this repository on your system.

4. Restart Claude Desktop

After restarting, Claude will have access to the following tools:

• run_fit - Run reflectivity fits on data
• assess_partial_data - Assess partial data quality
• extract_eis_intervals - Extract EIS timing intervals
• list_available_data - List available data sets
• list_available_models - List available fitting models
• get_tool_help - Get help for specific tools
• list_tools - List all available tools

5. Test the integration:

In Claude Desktop, you can now ask:

• "List the available neutron reflectometry tools"
• "What data sets are available for analysis?"
• "Run a fit on data set 218281 using the cu_thf model"

Alternative: Run MCP server manually for testing:

# Using the entry point
analyzer-mcp

# Or using fastmcp directly
fastmcp run analyzer_tools/mcp_server.py

# Or as a Python module
python -m analyzer_tools.mcp_server

• Tool overview: analyzer-tools or analyzer-tools --list-tools
• Specific tool help: analyzer-tools --help-tool <tool_name>
• Workflows: analyzer-tools --workflows
• Interactive selection: python -c "from analyzer_tools.welcome import help_me_choose; help_me_choose()"
• Developer notes: See docs/developer_notes.md

1. Follow the test-driven development approach outlined in docs/developer_notes.md
2. Update the tool registry when adding new tools
3. Run tests: pytest
4. Update documentation

• Add more flexibility for modifying existing models
• Add co-refinement
• Add nested sampling analysis to compare models
• Add the ability to call time-resolved reduction
• Add the ability to fit and report on time-resolved data