Parra-Glideator is an innovative web application designed to help paraglider pilots find the perfect place and time to fly. Leveraging sophisticated machine learning and generative AI models, it recommends the best flying spots based on weather forecasts and historical flight conditions.
Meet Parra-Glideator, our charming, paragliding gladiator parrot who traded natural flight for a paraglider. Like him, every pilot faces uncertainty—weather conditions, location choice, or flight planning can become a daunting battle. Parra-Glideator is here to ensure you have the odds on your side.
🌤️ Fly smarter, safer, and with more confidence!
The project is currently in public beta and available at parra-glideator.com. Try it out, plan your next flight adventure, and help us refine this pilot-friendly tool!
docker-compose.dev.yml Local dev stack (API, Postgres, Redis, Celery, frontend)
agents/ Ground Crew (site discovery & resources), Chat assistant
analytics/ Notebooks, datasets, training pipeline
art/ Brand assets (Parra-Glideator!)
backend/ FastAPI API, MCP server, Celery workers, Dockerfiles (web & worker)
db/ dbt project building the analytics warehouse
frontend/ React + Leaflet single-page app
gfs/ Library for downloading & flattening NOAA GFS data
net/ PyTorch models + preprocessing (Glideator-Net)
scrapers/ Scrapy spiders for XContest & Paragliding Map
Deployment note: production runs on Render and Render is the source of truth for production configuration. The Docker Compose file
docker-compose.dev.ymlat the repository root is for local development only.
# clone & launch everything (API + DB + Worker + Web)
$ git clone https://github.com/janhelcl/glideator.git
$ cd glideator
$ docker-compose -f docker-compose.dev.yml up --build- API docs: http://localhost:8000/docs
- Frontend: http://localhost:3000
- MCP Server http://localhost:8000/mcp
Each core component can be run on its own. Follow the dedicated README in the corresponding folder for setup & usage details:
backend/README.md— FastAPI API, Celery worker, Docker Compose details, Render deployment notesfrontend/README.md— React single-page applicationdb/README.md— dbt analytics warehousescrapers/README.md— Scrapy project for flight & site datagfs/README.md— GFS data downloader & utilitiesnet/README.md— PyTorch model libraryanalytics/training/README.md— End-to-end training pipelineagents/ground_crew/README.md— Ground Crew: browser agents, validation, exports for site resourcesagents/chat/README.md— Parra-Glideator chat assistant
- Backend (
backend/) – FastAPI, MCP server, PostgreSQL, Celery, Redis, deployed on Render in production. - Frontend (
frontend/) – React 18, Material-UI, React-Leaflet, D3. - Warehouse (
db/) – Postgres + dbt (staging & mart models). - ML Library (
net/) – Neural Networks implemented in PyTorch. - Training (
analytics/training/) – WebDataset loaders, notebooks. - Weather (
gfs/) – Fetches & processes NOAA GFS GRIB2 files. - Scrapers (
scrapers/) – Flight & site data collection with Scrapy. - Ground Crew (
agents/ground_crew/) – Browser-use pipelines that discover and validate local site/club links, extract webcam & meteostation URLs, and export JSON for the Glideator API (export-resources→backend/app/data/site_resources.json). Supersedes the old Site Researcher agent. - Chat (
agents/chat/) – Conversational assistant for the product. - MCP Integration – Model Context Protocol server enabling AI assistants to access paragliding data through structured tools for site information, weather forecasts, trip planning, and more.
- Scrapers write raw flights & sites → Postgres.
dbttransforms them into clean mart tables.- Training notebooks export WebDataset shards.
- PyTorch models are trained & the best checkpoint is shipped to the API.
For the full deep-dive (maths included!) see analytics/training/README.md.