AI System Development for Citizen Request Classification and Routing
Requests arriving at government agencies are often directed to the wrong department initially — the responsible executor wastes time, the applicant gets "not our competence" response, deadlines slip. AI classifier solves routing task with 92-95% accuracy on first routing attempt.
Multi-Level Classification
Government agency classifiers have hierarchical structure: federal classifier, agency classifier, territorial. Request must be assigned to correct hierarchy level and specific executor.
class RequestClassification(BaseModel):
federal_rubric: str # federal classifier rubric
department_rubric: str # agency classifier rubric
responsible_unit: str # responding department
responsible_officer: str | None # specific executor (if determined)
territorial_scope: str # federal / regional / municipal
subject_area: str # brief topic for registration
requires_field_inspection: bool # field visit needed
other_agencies: list[str] # other agencies to involve
confidence: float
Classifier Training
Foundation — historical data: thousands of requests with topic-executor markup. Quality metrics:
- Accuracy@1: correct department first try — target 92%+
- Accuracy@3: correct department in top-3 — target 99%+
- False routing rate: % of redirected requests — target < 5%
Territorial Routing
Requests about infrastructure require determining territorial affiliation via geocoding and administrative unit mapping.
Handling Mixed Requests
Single request often contains multiple independent questions. System splits such requests into sub-requests and routes separately.
Document Management System Integration
Supported DMS: DELO, DIRECTUM, Docsvision, 1C:Document Management. System passes request with pre-filled registration cards.