The AI Comments feature is an automated analytical comment generation system powered by artificial intelligence, integrated into the 3Bee monitoring platform. The system analyses cartographic layers (NDVI, LST, CLC, protected areas, etc.), air quality data, and IoT device data (PollyX and Spectrum), producing multilingual technical interpretations.
The process uses state-of-the-art LLM models: Gemini 3 Pro for visual map analysis (vision model) and Gemini 3 Flash for textual comment generation. Comments are generated in Italian and automatically translated into all platform-supported languages.
Automation is orchestrated via an AWS Step Function that daily: identifies eligible sites (with automation_ai_comments feature enabled), enqueues requests on SQS FIFO, monitors completion status with 10-minute polling intervals (maximum 24-hour timeout), and sends summary emails to site owners.
The billing system charges 3 credits per generated comment, with pre-generation credit availability check and post-generation charge.
| Component | Detail |
|---|---|
| Map analysis model | Gemini 3 Pro (vision) |
| Comment model | Gemini 3 Flash |
| Orchestration | AWS Step Function (STANDARD) |
| Message queue | Amazon SQS FIFO |
| Max concurrency | 10 sites in parallel |
| Status polling | 10 minutes, max 144 iterations (24h) |
| Cost | 3 credits per comment |
| Retry | 3 attempts with 2× exponential backoff |
| Endpoint | Method | Description |
|---|---|---|
/{user_slug}/sites/{site_id}/ai_map_comment | POST | Generate AI comment for a single layer |
/{user_slug}/sites/{site_id}/ai_map_comments | POST | Generate AI comments for multiple layers |
Rate limit: 1 request per 10 minutes, 6 per hour, 12 per day.
| Code | Name | Provider | Type | Coverage | Availability |
|---|---|---|---|---|---|
3BE_OASAI_99 | 3Bee Oasi AI Comments | 3Bee / Google AI | api | Global | Continuous |
automation_ai_comments flag in the features tableAutomatic generation of analytical comments using LLM models. Phase 1: visual analysis of maps with Gemini 3 Pro (vision model). Phase 2: text comment generation with Gemini 3 Flash. Phase 3: automatic multilingual translation. Orchestration: AWS Step Function with SQS FIFO queue and 10-minute polling. Input: cartographic layer images, air quality data, IoT device data. Output: multilingual text comment. Cost: 3 credits/comment. Source: lambda-oasi platform.