Semantic Search for Casino Game Discovery

Problem

A leading online casino and sports betting operator, with a rapidly growing library of over 11,000 casino games, sought to enhance its player experience when searching and browsing casino games. Their casino game discovery was limited, with the existing search relying on exact matches for game titles, provider names, or predefined tags. This created a high-friction user experience, as players couldn’t search for games based on themes, features, or gameplay styles, leading to low engagement with the vast majority of the game catalog.

Specific Pain Points:

• Limited game discovery

Players could only find games if they knew the exact title or provider’s name. The number of games that were tagged was limited. A search for “games with fruits” or “games with Egyptian theme” would return zero results, forcing users to manually browse thousands of titles. 

• Poor catalog utilization

A large part of the casino library was left undiscovered and underutilized.

• Inability to handle thematic queries

The system could not understand abstract concepts. A player looking for an “adventurous” or “scary” game had no way to search for one.

Solution

Description of the Solution:
We engineered and deployed an advanced, hybrid search solution that leverages Large Language Models to deliver a superior game discovery experience. The core of the solution is a new search endpoint that understands the thematic intent of a player’s query, handles local language nuances like diacritics, and intelligently blends semantic results with business priorities. This allows players to search in natural language and receive highly relevant, context-aware results instantly.

Specific Steps Taken:

• AI-Powered Data Enrichment: We processed the client’s catalog of over 11,000 games. Using a powerful generative LLM, we automatically generated rich, detailed descriptions, themes, and other metadata (e.g., rows, reels) for each game based on its name and provider.
• Vectorization and Indexing: We created vector embeddings from a combination of the game name, provider, and the newly generated description using OpenAI’s large embedding model. These embeddings were then stored and indexed using OpenSearch.
• Hybrid Search Endpoint Development: The primary deliverable was a new search API endpoint. This endpoint intelligently combines multiple strategies:

• Step 1 (Keyword Search): It first utilizes the client’s existing, proven keyword search for direct matches.
• Step 2 (Semantic Search): It then queries OpenSearch to find semantically similar games based on the user’s query.
• Business Logic: The results are then ranked, taking into account sponsorships and promoted providers to meet business objectives.
• Diacritics Handling: The search logic was built to be robust against missing diacritics, ensuring queries are successful even with common typing shortcuts.

• Automated Relevancy Evaluation: To ensure quality, we developed an innovative evaluation framework. An LLM was used to automatically rate the relevancy of search results against a variety of test queries, allowing for rapid, data-driven tuning of the search algorithm.
• Real-Time Indexing via Kafka: We implemented a Kafka consumer to listen for events when games are added or modified. This consumer automatically triggers the AI data enrichment and embedding process, adding new games to the search index so they become searchable almost instantly after being added to the platform.

Unique Value Proposition:

The solution’s unique value lies in its intelligent, hybrid search capability. By blending traditional keyword matching with advanced semantic understanding, the platform can interpret a player’s natural language and true intent. This transforms the casino lobby from a static grid into a dynamic discovery engine, making the entire 11,000+ game catalog accessible and setting the client apart from competitors who rely on basic filtering.

Results

The implementation of the semantic search solution led to several positive outcomes for the client:

• Improved Discoverability: Users experienced an enhanced ability to find relevant games, resulting in increased engagement and satisfaction with the platform.
• Higher Search Relevance: The hybrid search approach, combined with LLMs, ensures that users receive more accurate and contextually appropriate results, improving the overall search experience.
• Future-Proof Scalability: The solution was designed to be scalable, allowing the client to expand its massive game database further without compromising search performance.
  

Smart Tip

Regardless of the industry—from e-commerce and gaming to enterprise knowledge bases—implementing semantic search is a powerful business driver. By understanding user intent rather than just keywords, it delivers a superior experience, directly boosting engagement, customer satisfaction, and key performance metrics.

Smart Fact

Using an LLM for automated result evaluation can drastically speed up the development cycle. It allows for thousands of queries to be tested for relevancy in minutes, a task that would take human testers days to complete.

About the Clients

The client is a major international online gaming group founded in the early 2000s. With a strong presence in numerous countries, they are known for their comprehensive sportsbook and a rich portfolio of online casino games, focusing on technological innovation to enhance the user experience.

Technologies Used

• Embedding Models: OpenAI Large Embedding Model
• Generative LLMs: GPT-4.1 (for data enrichment & evaluation)
• Vector Database / Search: OpenSearch
• Messaging System: Apache Kafka
• Programming Languages: Python, Java, JavaScript