Development of an AI system for e-Government AI
E-government has accumulated vast amounts of data on citizen requests, administrative processes, and services provided. AI accelerates the processing of this data and makes citizen interaction with the government easier.
Automation of public service delivery
Intelligent Citizen Assistant:
LLM + RAG based on government services regulations:
from langchain.vectorstores import Chroma
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.chains import RetrievalQA
from langchain.chat_models import ChatOpenAI
def create_govservices_assistant(regulations_db_path):
"""
Ассистент по госуслугам с RAG на базе административных регламентов.
"""
embeddings = HuggingFaceEmbeddings(
model_name='intfloat/multilingual-e5-large'
)
vectorstore = Chroma(
persist_directory=regulations_db_path,
embedding_function=embeddings
)
retriever = vectorstore.as_retriever(
search_kwargs={"k": 5, "score_threshold": 0.75}
)
llm = ChatOpenAI(model='gpt-4o-mini', temperature=0)
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=retriever,
return_source_documents=True,
chain_type_kwargs={
"prompt": GOVSERVICE_PROMPT_TEMPLATE
}
)
return qa_chain
Typical queries: - "What documents are needed to register as an individual entrepreneur?" → list with links to regulations - "My SNILS is lost, what should I do?" → step-by-step instructions - "How much is the state fee for a new-style passport?" → up-to-date information
Automation of document flow
OCR + NLP for incoming documents:
Citizens and organizations submit documents in various formats. Automatic processing: - OCR: Tesseract / EasyOCR / PaddleOCR for scans - NLP extraction: name, date of birth, SNILS, TIN, address — from free text - Validation: SNILS/TIN check digits, format checks - Routing to the correct executor based on the classification of the request type
Automatic preparation of answers:
For typical requests with a clear answer: - Income statement, extract from the register - generated from the database using a template - 80% of typical requests can be closed automatically without the involvement of an inspector
Detecting fraud and violations
Social benefits:
ML detection of illegal receipt of benefits: - Double receipt of subsidies based on different documents (passport series/number + SNILS reconciliation) - Receipt of unemployment benefits despite active employment (cross-check with Federal Tax Service + Pension Fund) - Anomalies in address data: one address for 50+ benefit recipients
Tax risks:
Automatic scoring of taxpayers based on the risk of underestimating the tax base: - Indicators: turnover vs. tax burden by industry (benchmark) - Anomalies in the transaction structure - Links to known tax schemes through the company graph
Queue and recording management
Predictive load management of MFC:
- Attendance forecast by service type, day of the week, period (quarterly peaks) - Dynamic window management: open an additional window if there is a long queue - Online booking: ML selects slots with minimal load
SMS/Push notifications:
- Application status → automatic notifications at every step - Proactive: "Your police clearance certificate is ready. Pick it up by January 15" - Reminders: "Your driver's license expires in 30 days"
Analytics and Politics
Monitoring the quality of public services:
- NLP analysis of citizen reviews on Gosuslugi and social media - Sentiment for each agency and service type - Diagram of complaint reasons → where systemic problems lie
Budget planning:
ML forecast of demand for social benefits for budget planning: - Forecast of the number of pensioners, unemployed, beneficiaries with a 3-5 year horizon - Scenario analysis: what will happen if the criteria for assignment change
Development time: 4–7 months for an e-Government AI system with an LLM assistant, document automation, and anti-fraud analytics.