AI Public Procurement System Development

We design and deploy artificial intelligence systems: from prototype to production-ready solutions. Our team combines expertise in machine learning, data engineering and MLOps to make AI work not in the lab, but in real business.
Showing 1 of 1 servicesAll 1566 services
AI Public Procurement System Development
Complex
~2-4 weeks
FAQ
AI Development Areas
AI Solution Development Stages
Latest works
  • image_web-applications_feedme_466_0.webp
    Development of a web application for FEEDME
    1161
  • image_ecommerce_furnoro_435_0.webp
    Development of an online store for the company FURNORO
    1041
  • image_logo-advance_0.png
    B2B Advance company logo design
    561
  • image_crm_enviok_479_0.webp
    Development of a web application for Enviok
    823
  • image_logo-aider_0.jpg
    AIDER company logo development
    762
  • image_crm_chasseurs_493_0.webp
    CRM development for Chasseurs
    848

Development of an AI system for public procurement Public Procurement AI

Public procurement is a sector with a high level of abuse: cartels, supplier affiliations, and inflated maximum contract prices. The ML system analyzes procurements under the Unified Information System (UIS) and identifies signs of violations that are not detectable through manual audits.

Data and sources

Procurement information base:

data_sources = {
    'eis_zakupki_gov_ru': {
        'api': 'Открытые данные ЕИС API (44-ФЗ, 223-ФЗ)',
        'entities': ['ContractNotice', 'ContractAward', 'Supplier', 'OKPD2'],
        'volume': '~4 млн закупок в год'
    },
    'egrul_fns': {
        'source': 'ЕГРЮЛ ФНС — данные о юрлицах',
        'use': 'связи между поставщиками, аффилированность, учредители'
    },
    'rosstat': {
        'source': 'финансовая отчётность компаний',
        'use': 'реальные возможности поставщика vs. объём контракта'
    },
    'sudrf_ru': {
        'source': 'арбитражные дела',
        'use': 'история судебных споров поставщиков'
    }
}

Cartel Detection

Signs of a cartel in trading:

import pandas as pd
import numpy as np
from itertools import combinations

def detect_collusion_in_auction(auction_bids: pd.DataFrame,
                                 auction_id: str) -> dict:
    """
    Классические паттерны картеля:
    1. Cover bidding: один выигрывает, остальные делают заведомо проигрышные ставки
    2. Bid suppression: конкуренты отказываются от участия
    3. Bid rotation: участники по очереди побеждают в разных лотах
    4. Market allocation: географическое или секторальное разделение
    """
    bids = auction_bids[auction_bids['auction_id'] == auction_id]

    if len(bids) < 2:
        return {'status': 'insufficient_bidders'}

    nmck = bids['start_price'].max()  # НМЦК — начальная (максимальная) цена

    # Снижение цены каждого участника
    bids['reduction_pct'] = (nmck - bids['bid_price']) / nmck * 100
    bids_sorted = bids.sort_values('bid_price')

    # Признак 1: минимальный шаг снижения от предыдущего участника
    # Cover bids: остальные снижаются ровно на 0.5% (разрешённый минимум)
    reductions = bids_sorted['bid_price'].diff().abs() / bids_sorted['bid_price'].shift(1)
    min_step_count = (reductions < 0.006).sum()  # почти все снизились на минимум

    # Признак 2: победитель снизил на большой %, остальные — на минимум
    winner_reduction = bids_sorted.iloc[0]['reduction_pct']
    losers_reductions = bids_sorted.iloc[1:]['reduction_pct']
    reduction_gap = winner_reduction - losers_reductions.mean()

    # Признак 3: временные паттерны подачи заявок
    if 'bid_timestamp' in bids.columns:
        time_deltas = bids['bid_timestamp'].sort_values().diff().dt.total_seconds()
        very_fast_submissions = (time_deltas < 60).sum()  # < 1 минуты между заявками
    else:
        very_fast_submissions = 0

    collusion_indicators = {
        'cover_bids': min_step_count >= len(bids) - 2,
        'large_reduction_gap': reduction_gap > 10,
        'fast_sequential_bids': very_fast_submissions > 1,
        'few_participants': len(bids) <= 2
    }

    collusion_score = sum(collusion_indicators.values()) / len(collusion_indicators)

    return {
        'auction_id': auction_id,
        'n_bidders': len(bids),
        'winner_reduction_pct': round(winner_reduction, 2),
        'collusion_indicators': collusion_indicators,
        'collusion_score': round(collusion_score, 2),
        'flag_for_review': collusion_score > 0.5
    }

Supplier Affiliation Analysis

Register Relationship Graph:

import networkx as nx

def build_supplier_affiliation_graph(suppliers: pd.DataFrame,
                                      egrul_data: pd.DataFrame) -> nx.Graph:
    """
    Связи: общие учредители, адреса, телефоны, директора.
    Аффилированные компании = один бенефициар = конкурент только формально.
    """
    G = nx.Graph()

    for _, supplier in suppliers.iterrows():
        G.add_node(supplier['inn'], type='supplier', name=supplier['company_name'])

    # Связи через общих учредителей
    founders_data = egrul_data.groupby('founder_inn')['company_inn'].apply(list)
    for founder, companies in founders_data.items():
        if len(companies) > 1:
            for c1, c2 in combinations(companies, 2):
                if G.has_node(c1) and G.has_node(c2):
                    G.add_edge(c1, c2, relation='common_founder', founder_inn=founder)

    # Связи через адрес регистрации (массовые регистрации)
    address_counts = egrul_data.groupby('legal_address')['company_inn'].apply(list)
    for address, companies in address_counts.items():
        if len(companies) > 5:  # адрес массовой регистрации
            for c1, c2 in combinations(companies, 2):
                if G.has_node(c1) and G.has_node(c2):
                    G.add_edge(c1, c2, relation='shared_address')

    return G

def find_affiliated_bidders(auction_bidders: list, affiliation_graph: nx.Graph) -> dict:
    """
    Проверяем: есть ли прямая связь между участниками?
    """
    affiliated_pairs = []
    for i, bidder1 in enumerate(auction_bidders):
        for bidder2 in auction_bidders[i+1:]:
            if affiliation_graph.has_edge(bidder1, bidder2):
                path = nx.shortest_path(affiliation_graph, bidder1, bidder2)
                affiliated_pairs.append({
                    'bidder1': bidder1,
                    'bidder2': bidder2,
                    'connection_path': path,
                    'connection_type': affiliation_graph[bidder1][bidder2].get('relation')
                })

    return {
        'affiliated_pairs': affiliated_pairs,
        'has_affiliation': len(affiliated_pairs) > 0,
        'risk_level': 'high' if len(affiliated_pairs) > 0 else 'low'
    }

Analysis of the initial maximum contract price (price justification)

Detection of elevated NMCK:

from sklearn.ensemble import GradientBoostingRegressor

def detect_inflated_nmck(procurement: dict, similar_procurements: pd.DataFrame) -> dict:
    """
    НМЦК должна отражать рыночную стоимость.
    Сравниваем с аналогичными закупками: тот же ОКПД2, регион, объём.
    """
    # Фильтрация аналогов
    comparable = similar_procurements[
        (similar_procurements['okpd2_code'].str[:4] == procurement['okpd2_code'][:4]) &
        (similar_procurements['region_code'] == procurement['region_code']) &
        (similar_procurements['quantity'] >= procurement['quantity'] * 0.5) &
        (similar_procurements['quantity'] <= procurement['quantity'] * 2.0)
    ]

    if len(comparable) < 5:
        return {'status': 'insufficient_comparables'}

    unit_prices = comparable['contract_price'] / comparable['quantity']
    expected_unit_price = unit_prices.median()
    std_unit_price = unit_prices.std()

    nmck_unit = procurement['nmck'] / procurement['quantity']
    z_score = (nmck_unit - expected_unit_price) / (std_unit_price + 1e-9)

    overpricing_pct = (nmck_unit - expected_unit_price) / expected_unit_price * 100

    return {
        'nmck_unit_price': round(nmck_unit, 2),
        'market_median_price': round(expected_unit_price, 2),
        'overpricing_pct': round(overpricing_pct, 1),
        'z_score': round(z_score, 2),
        'inflation_detected': z_score > 3,
        'comparable_contracts_n': len(comparable)
    }

Bid Rotation – winning pattern

Statistical Test for Winner Rotation:

from scipy.stats import chi2_contingency

def detect_bid_rotation(procurement_history: pd.DataFrame,
                          supplier_group: list) -> dict:
    """
    Картельная ротация: каждый поставщик выигрывает "свою" долю лотов.
    Статистически: равномерное распределение побед при видимой конкуренции.
    """
    group_wins = procurement_history[
        procurement_history['winner_inn'].isin(supplier_group)
    ]['winner_inn'].value_counts()

    if len(group_wins) < 2:
        return {'status': 'insufficient_data'}

    # Chi-square тест: равномерность распределения побед
    observed = group_wins.values
    expected = np.full_like(observed, np.mean(observed), dtype=float)
    chi2_stat, p_value = chi2_contingency([observed, expected])[:2]

    # Низкий chi2 = слишком равномерное распределение = подозрительно
    rotation_detected = chi2_stat < 2 and p_value > 0.9

    return {
        'supplier_win_counts': group_wins.to_dict(),
        'chi2_statistic': round(chi2_stat, 3),
        'p_value': round(p_value, 3),
        'rotation_detected': rotation_detected,
        'interpretation': 'Слишком равномерное распределение побед' if rotation_detected else 'Распределение нормальное'
    }

Integration with regulators: Export of detected violations in the formats of the Federal Antimonopoly Service of Russia (Anti-Cartel Department), the Accounts Chamber, and Rosfinmonitoring. REST API for integration with automated procurement control systems (ASFK, ACC-Finance).

Deadlines: Basic collision analysis + initial maximum contract price comparison + affiliation according to the Unified State Register of Legal Entities — 5-6 weeks. Affiliation graph, bid rotation test, ML model for corruption indicators, FAS API — 3-4 months.