How elouansalaun/lidar_agent works
LiDAR AI Agent answers Montreal-area questions by loading and processing airborne point clouds, classifying points, extracting and filtering buildings, validating tool results, and presenting answers with maps and visualizations. Users can run this through the agent or a fixed-parameter pipeline UI; the repository also includes a model-training and evaluation path. The README documents the main pipeline and agent capabilities; some subsystem links below rely on that documentation and the SOURCE INDEX rather than sampled implementation excerpts.
Main entry points: registry.py Tool registry · streamlit_app.py Streamlit interface · graph.py Agent graph · lidar_tools.py LiDAR tools
Components 23
User interfaces
- Streamlit interface
streamlit_app.py - Fixed workflow
src/workflow.py
Agent and orchestration
- Agent graph
src/agent/graph.py - LLM adapter
src/agent/llm.py - Offline planner
src/agent/offline.py - Result validator
src/agent/validator.py - Tool registry
src/tools/registry.py
LiDAR processing
- LiDAR tools
src/tools/lidar_tools.py - Point-cloud I/O
src/lidar/io.py - Ground normalization
src/lidar/ground.py - Feature extraction
src/lidar/features.py
Classification and buildings
- Point classifier
src/ml/classify.py - Classification tools
src/tools/ml_tools.py - Training dataset
src/ml/dataset.py - Building analysis
src/buildings/analysis.py - GIS tools
src/tools/gis_tools.py
Visualization and data
- Interactive maps
src/visualization/maps.py - Result views
src/visualization/figures.py - Raw LiDAR tiles
data/raw_tiles - Processed data
data/processed - Generated outputs
data/outputs
Other components
- Questioner
- Claude API
Connections 27
- Questioner → Streamlit interface: asks question
README.md - Streamlit interface → Agent graph: runs agent
streamlit_app.py - Streamlit interface → Fixed workflow: runs pipeline mode
README.md - Agent graph → LLM adapter: requests plan
src/agent/graph.py - LLM adapter → Claude API: calls API
src/agent/llm.py - LLM adapter → Offline planner: selects fallback
src/agent/llm.py - Agent graph → Tool registry: dispatches tools
src/agent/graph.py - Agent graph → Result validator: validates results
src/agent/graph.py - Fixed workflow → Tool registry: runs tools
src/workflow.py - Fixed workflow → Result validator: checks results
src/workflow.py - Tool registry → LiDAR tools: registers tools
src/tools/lidar_tools.py - Tool registry → Classification tools: registers tools
src/tools/ml_tools.py - Tool registry → GIS tools: registers tools
src/tools/gis_tools.py - LiDAR tools → Point-cloud I/O: loads and saves
src/tools/lidar_tools.py - Point-cloud I/O → Raw LiDAR tiles: reads tiles
README.md - LiDAR tools → Ground normalization: normalizes height
src/tools/lidar_tools.py - LiDAR tools → Feature extraction: preprocesses points
- Classification tools → Point classifier: classifies points
src/tools/ml_tools.py - Point classifier → Feature extraction: computes features
src/ml/classify.py - GIS tools → Building analysis: extracts buildings
src/tools/gis_tools.py - GIS tools → Interactive maps: creates map
src/tools/gis_tools.py - Building analysis → Interactive maps: supplies footprints
README.md - Training dataset → Feature extraction: builds samples
src/ml/dataset.py - Training dataset → Point-cloud I/O: loads blocks
src/ml/dataset.py - Point-cloud I/O → Processed data: writes intermediates
- Interactive maps → Generated outputs: saves map
README.md - Streamlit interface → Result views: renders results
streamlit_app.py