ConstructionSaaS · Web application · AI/ML
AI project scheduling app rescue
We rescued an AI-assisted project scheduling product 48 hours before a customer demo: rebuilt an undocumented codebase locally, fixed six core Gantt features, persisted user settings, and kept the AI chat working through a model-provider outage.
Problem
The client had inherited a codebase with no documentation, a manual build process and broken core features, and had a customer demo within days.
Solution
The team rebuilt the development setup without documentation, then fixed saving in the review screen, task dependencies, percentage-complete shading in bars, start-date propagation, and column width and show/hide on the grid, persisting hidden columns and workweek, holiday and working-hour settings with the file. When the chat feature failed during a model-provider capacity incident, the integration was switched to another provider through LangChain for the demo. Changes were reviewed live with the client and pushed incrementally.
Key features
- Local build of an undocumented Angular front end and Python back end
- Save in the review screen
- Task dependencies in review and generate screens
- Percentage-complete shading in Gantt bars
- User start date propagated to the generated chart
- Adjustable column widths and dynamic show/hide columns
- Hidden columns and schedule settings persisted with the saved file
- AI chat stabilised by switching model provider during an outage
Tech stack
- Angular
- Anthropic
- Claude Sonnet
- GitHub
- LangChain
- OpenAI
- Python
- Syncfusion
- TypeScript
- langchain_openai