AI System for Automating Citizen Request Processing

Your department receives 10,000+ requests per month—from housing and utilities inquiries to complaints about official actions. Manual processing of each takes 3–5 days, and missing deadlines under Federal Law No. 59-FZ risks fines and complaints to the prosecutor's office. The situation is compounded by a 20% annual increase in requests and template inquiries that consume 80% of employee time. We develop AI systems that automate reception, classification, routing, and response preparation, reducing processing time by 5–10 times. Our experience spans over 10 years in AI and 50+ implementations in the public sector. Get an analysis of your stream in 2 days—contact us.

AI System for Automating Citizen Request Processing: How It Works

Key System Modules

Reception and Integration with Government Resources

The system integrates with ESIA, SMEV, email, and the portal. It normalizes data into a single format. Through ESIA, it obtains verified applicant data (full name, SNILS, address). SMEV allows automatic interagency requests—for example, data from the Rosreestr for land-related inquiries. Integration with GIS Housing and Utilities and EPGU (Gosuslugi) is also available for status publication.

Classification and Data Extraction

from pydantic import BaseModel
from enum import Enum

class RequestCategory(str, Enum):
    HOUSING = "жилищные вопросы"
    UTILITIES = "ЖКХ"
    LAND = "земельные отношения"
    SOCIAL = "социальная защита"
    TRANSPORT = "транспорт и дороги"
    ECOLOGY = "экология"
    PERMISSIONS = "разрешения и лицензии"
    COMPLAINT = "жалоба на действия должностных лиц"
    OTHER = "прочее"

class CitizenRequest(BaseModel):
    applicant_name: str
    applicant_contact: str
    request_text: str
    attachments: list[str]

class ProcessedRequest(BaseModel):
    category: RequestCategory
    subcategory: str
    subject_summary: str          # краткое изложение в 1-2 предложениях
    responsible_department: str
    priority: str                 # routine / urgent / special_control
    deadline_days: int            # расчётный срок ответа по 59-ФЗ
    requires_interdepartmental: bool  # нужен ли межведомственный запрос
    extracted_addresses: list[str]    # адреса из текста обращения
    extracted_organizations: list[str]
    is_repeat: bool               # повторное обращение
    related_request_ids: list[str]

def process_citizen_request(request: CitizenRequest, db) -> ProcessedRequest:
    # Поиск похожих предыдущих обращений
    similar = db.semantic_search(request.request_text, top_k=5)

    context = build_context(similar)
    return llm.parse(
        build_classification_prompt(request, context),
        response_format=ProcessedRequest
    )

Deadline Calculation per 59-FZ

According to 59-FZ, the base deadline for request processing is 30 days, extendable by 30 days for interagency requests. The calculation is non-trivial: exceptions exist—housing and utilities requests may require a 10-day response per regional regulations, urgent requests—15 days. An interagency request extends the deadline by 30 days with notification to the applicant.

def calculate_deadline(
    request: ProcessedRequest,
    received_date: date,
    holiday_calendar: HolidayCalendar
) -> DeadlineInfo:

    base_days = 30  # базовый срок по 59-ФЗ ст. 12

    if request.priority == "urgent":
        base_days = 15
    elif request.category == RequestCategory.UTILITIES:
        base_days = 10  # региональные требования

    if request.requires_interdepartmental:
        base_days += 30  # ст. 10 ч. 2 59-ФЗ

    # Рабочие дни с учётом производственного календаря
    deadline = holiday_calendar.add_working_days(received_date, base_days)

    return DeadlineInfo(
        deadline=deadline,
        warning_date=holiday_calendar.add_working_days(received_date, base_days - 5),
        escalation_date=holiday_calendar.add_working_days(received_date, base_days - 2)
    )

Generation of Draft Responses

For standard requests (80–90% of the incoming stream), the system automatically generates a draft response. The response includes references to regulatory legal acts (NPAs) and specific explanations, not generic phrases. Compare: manual preparation takes 2–4 hours, AI generation takes 10–15 minutes with 95% accuracy.

def generate_draft_response(
    request: ProcessedRequest,
    original_text: str,
    knowledge_base: KnowledgeBase
) -> DraftResponse:

    # Поиск релевантных НПА, постановлений, регламентов
    relevant_docs = knowledge_base.search(
        query=original_text,
        doc_types=["law", "regulation", "instruction", "precedent"],
        top_k=10
    )

    # Генерация ответа со ссылками
    prompt = f"""Подготовь официальный ответ на обращение гражданина.

Обращение: {original_text}
Тематика: {request.category}

Релевантные НПА:
{format_documents(relevant_docs)}

Требования:
- Официальный деловой стиль
- Конкретные ссылки на статьи нормативных актов
- Описание порядка действий для заявителя
- Без общих фраз и отписок"""

    draft = llm.generate(prompt, max_tokens=800)

    return DraftResponse(
        text=draft,
        referenced_documents=[d.id for d in relevant_docs[:5]],
        confidence=estimate_confidence(request, relevant_docs),
        requires_human_review=request.priority == "urgent" or request.category == RequestCategory.COMPLAINT
    )

Why HDBSCAN for Clustering?

Detecting systemic issues requires a noise-robust algorithm. HDBSCAN does not require specifying the number of clusters and identifies outliers, which is critical for real data. Example:

def detect_systemic_issues(
    requests: list[ProcessedRequest],
    period_days: int = 30
) -> list[SystemicIssue]:

    # Кластеризация по тематике и адресам
    clusterer = HDBSCANClusterer(min_cluster_size=10)
    clusters = clusterer.fit(requests)

    issues = []
    for cluster in clusters:
        if cluster.growth_rate > 2.0:  # рост числа обращений в 2+ раза
            issues.append(SystemicIssue(
                category=cluster.dominant_category,
                location=cluster.most_common_address,
                request_count=len(cluster.requests),
                sample_texts=cluster.get_samples(n=3),
                trend="growing",
                recommended_action=generate_action_recommendation(cluster)
            ))

    return sorted(issues, key=lambda x: x.request_count, reverse=True)

Comparison of Manual and AI Processing

AI classification is 15–25 percentage points more accurate than manual, and response generation speed is 12–16 times higher. Budget savings for a department can reach 5 million rubles per year at a stream of 10,000 requests. Get a detailed savings calculation for your department.

How AI Reduces Request Processing Time by 5–10 Times?

Thanks to automatic classification and response generation for 80–90% of standard requests. Employees handle only complex and non-standard inquiries. The system automatically tracks deadlines and escalates overdue items.

What's Included in the Pilot Implementation?

What's Included in the Final Deliverable

• Documentation: technical documentation, operator instructions, administrator guide.
• Access: to system API, dashboards, logs.
• Training: training for 10–15 employees, self-paced materials.
• Support: 3 months of warranty support, SLA for incidents.
• Source code: classification and response generation modules (optional).

Our Experience and Guarantees

We have been working with AI solutions for over 10 years, implementing 50+ systems in the public sector. Key projects:

• Automation of request processing for a regional ministry (70% time reduction, 3.5 million rubles annual savings).
• Deadline control system for a federal agency (95% reduction in fines).

We guarantee: compliance with 59-FZ, certified security, phased implementation without interrupting current operations. Request a consultation—we'll analyze your stream in 2 days for free.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.