📝 Description
FireForm currently relies primarily on the user’s incident narrative and the LLM extraction pipeline to populate report fields.
For many emergency and wildfire reports, however, some fields should not be guessed or inferred by the LLM when they can be retrieved from trusted deterministic sources.
Examples:
- wind speed at incident time
- humidity
- temperature
- geographic coordinates
- jurisdiction / county / district
- incident location validation
This issue proposes adding a trusted context enrichment layer between initial incident extraction and final PDF filling.
The goal is to let the LLM focus on understanding the responder’s narrative, while deterministic provider modules enrich objective facts such as weather and geography.
Conceptually:
Incident text / voice transcript
↓
LLM extracts narrative fields
↓
Context enrichment layer
↓
Weather provider + geocoding / jurisdiction provider
↓
Validated enriched incident JSON
↓
Template mapping + PDF fill
📝 Description
FireForm currently relies primarily on the user’s incident narrative and the LLM extraction pipeline to populate report fields.
For many emergency and wildfire reports, however, some fields should not be guessed or inferred by the LLM when they can be retrieved from trusted deterministic sources.
Examples:
This issue proposes adding a trusted context enrichment layer between initial incident extraction and final PDF filling.
The goal is to let the LLM focus on understanding the responder’s narrative, while deterministic provider modules enrich objective facts such as weather and geography.
Conceptually: