What enters the process
Collection events and alarm events generated by vehicle-camera systems and received by the existing platform.
Waste management technology
AI filtering for vehicle-camera events in a waste management platform
An ongoing client engagement for an existing waste management platform that receives events from vehicle-camera systems. DapperAgent added an AI-assisted filtering component to that platform.
- Client
- An anonymized waste management technology platform
- DapperAgent contribution
- An AI-assisted component that evaluates visual evidence and filters selected event types according to configured rules inside the existing platform workflow.
- Engagement type
- Implementation with ongoing monitoring, support, and optimization
01
Original workflow and problem
How the work was handled
Vehicle-camera systems generate events that enter the existing platform workflow.
What the automation was intended to improve
Add a filtering step that can identify likely false events and apply configured rules to selected event types.
02
Before and after
Before
Vehicle-camera systems generate events that enter the existing platform workflow.
After
An additional AI filtering step evaluates visual evidence and applies configured rules to selected event types.
03
What we built
AI analysis
Evaluate visual evidence
HuggingFaceTB/SmolVLM2-500M-Video-Instruct analyzes the visual evidence used by the filtering component to identify likely false events.
Software rules
Apply configured filtering rules
The component applies configured decision rules to selected event types and integrates that step into the existing platform workflow.
Operational boundary
Extend the platform, not replace it
DapperAgent added the filtering component. The cameras, original camera event detector, and waste management platform were already in place.
04
What changed in the workflow
- Selected vehicle-camera events can be evaluated against visual evidence before continuing through the platform workflow.
- Likely false events can be identified by the added filtering component.
- Filtering behavior is applied through configured rules for selected event types.
05
Technical implementation
The filtering component integrates HuggingFaceTB/SmolVLM2-500M-Video-Instruct for video analysis. The model supports the visual evaluation step; conventional integration and configured rules control how selected event types are handled inside the existing platform.
06
Ongoing operation
The engagement continues with monitoring, support, and optimization of the filtering component and its integration.