Bussol Smart Search Ultra — AI search for an online store

What this solution does

Bussol SmartSearch Ultra is an intelligent search module for online stores built on the 1C-Bitrix platform. The solution combines classic full-text search, morphological analysis, synonym handling, relevance management, artificial intelligence capabilities, user behavior analytics, voice input, and hybrid image-based product search. The module is designed for stores where standard search is insufficient: customers make typos, use natural language, and search by brand, price, specifications, availability, and images. Store owners need to understand what users are looking for, which results attract clicks, and which queries lead to orders.

Key features include fast full-text search for large catalogs using a FULLTEXT index. The system accounts for typos, keyboard layouts, morphology, and word forms. A user-defined synonym dictionary is available for terms, brands, categories, and colloquial names. Rewrite rules allow the system to search for Y when a user enters X. Stop words filter out insignificant terms from search queries. Field weights are configurable for product name, description, properties, article number, XML_ID, and other product data.

The Bussol SmartSearch Ultra module allows setting priorities for information blocks, sections, and specific products. Products in stock are automatically boosted in the results, while products without a price or stock are excluded. Independent ranking by price, availability, and popularity is implemented.

Hybrid semantic search is supported, combining standard text relevance with AI embeddings. This enables finding products not only by exact word matches but also by the meaning of the query. For example, a user can describe a product in words without knowing the exact category name or model. To accelerate processing, a local ANN/LSH index is used to pre-select the most probable candidates, after which the module performs precise vector comparison.

The module can extract commercial conditions and product characteristics from search phrases. Example queries: Samsung smartphone under 50000 in stock, Lenovo laptop from 70000 to 120000, red dress size 46, 55-inch TV under 100000. The system automatically determines the minimum and maximum price, brand, availability, and values of indexed product properties from the query.

Bussol SmartSearch Ultra is a module for online stores featuring a configurable AI layer. The administrative panel allows selection of a provider and specific model. Supported options include Jina AI for text and multimodal embeddings, Cohere for text and image embeddings in supported models, OpenAI for text embeddings enabling semantic text search, and a custom AI endpoint via the Custom / Unified JSON format. The Legacy option ensures compatibility with the previous embedding endpoint scheme. AI functions are optional; if a provider is temporarily unavailable, standard search continues to operate without interruption.

The module supports two-level image-based product search. The first level uses a local perceptual hash that operates without external neural networks. The second level employs image embeddings for more intelligent comparison. A local ANN index accelerates searching through embedding vectors. Hybrid ranking combines results from perceptual hash and semantic similarity. If the embedding provider is disabled or unavailable, the system automatically switches to searching via the local perceptual hash.

The SmartSearch component supports voice input through the Web Speech API in compatible browsers, allowing users to dictate queries instead of typing them manually. Results are displayed on a full-fledged results page.

The Bussol SmartSearch Ultra module for online stores provides not only a fast dropdown search but also a dedicated results page component. This page supports pagination and allows sorting products by relevance, price, popularity, and availability. The system displays the total count of found items.

Administrators can access relevance diagnostics that explain why a specific product appeared at a certain position. The tool shows the contribution of individual factors: text fields (including NAME), properties, morphology, semantic proximity, priorities, availability, price, popularity, and the self-learning ranking mechanism. This enables search tuning based on objective data rather than subjective assumptions.

The module collects detailed search analytics to analyze user interaction with the system. Reports include statistics on search queries and zero-result queries, the click-through rate (CTR) of results, user search history, and the 'chasto ishchut' (chasto ishchut) block. Additional data covers conversion analytics, the link between search queries and subsequent orders, the search-to-purchase coefficient, and revenue generated through search.

The module utilizes accumulated statistics on impressions, clicks, and purchases as an additional ranking signal. The self-learning mechanism does not replace text relevance but precisely boosts the positions of products that demonstrate higher conversion rates in specific search scenarios.

A/B testing is implemented to evaluate the quality of the trainable ranking. Users are consistently distributed between a control group and an experimental group. This allows for comparison of search query volume, CTR, conversion counts, conversion rate, and total revenue for each group.

Import and export support for CSV files enables the transfer of settings and mass editing. CSV files can be used to manage synonyms, rewrite rules, and the priority levels of information blocks, sections, and individual products.

The bussol:smartsearch component ensures fast result retrieval. The interface supports voice input, image search, query history display, and a 'Frequently Searched' block. A search bar clear button is available. The color of the 'Search' button, the button text color, and the search bar border color are independently configurable. In a multisite configuration, settings are saved separately for each site.

The module provides standard security measures.

The current version enhances security for public and administrative actions. CSRF protection is implemented for state-changing operations. Session ownership is verified when click events are recorded. A signed visitor ID is used for anonymous users. HttpOnly and SameSite cookie attributes are configured. The custom AI endpoint is protected against SSRF attacks. Requests to localhost and private or reserved IP addresses are blocked for external AI queries. Secret API keys are not displayed in the administrative form. Restrictions have been applied to image processing.

SmartSearch Ultra is particularly beneficial for online stores with large catalogs, numerous properties and attributes, and products from various brands and SKUs. It is ideal for scenarios where standard search often returns zero or irrelevant results, where search behavior analytics are critical, where image search is required, or where AI search must be implemented without replacing the entire catalog infrastructure.

The core principle of SmartSearch Ultra is that AI augments search rather than making the store dependent on AI.

The Bussol Smart Search Ultra module provides core search functions: morphology, synonym handling, full-text search, perceptual hashing, commercial ranking, and analytics. These components can operate independently. The AI embeddings and semantic search layer can be connected as an additional quality layer.

If “Bussol Smart Search Ultra — AI search for an online store” fits the task, General iT can install the module, configure it and verify the result in the project.