How personalization and recommendations enhance search and discovery

## 🚀 Meilisearch 1.12 release

![Meilisearch 1.12 release notes](https://unable-actionable-car.media.strapiapp.com/meilisearch_1_12_7139b280d2.jpg)

Meilisearch 1.12 release notes
Meilisearch 1.12 significantly **improves indexing speed** and introduces new APIs to **disable prefix search and facet search** and retrieve **task batches**.

[Learn more](https://www.meilisearch.com/blog/meilisearch-1-12)

## 💫 Recent releases wrap up 

![Group 143726039 (2).png](https://unable-actionable-car.media.strapiapp.com/Group_143726039_2_a316baf38d.png)

Catch up on everything we shipped recently—faster search, revamped search preview, new index settings UI, and more.

[Learn more](https://www.meilisearch.com/blog/q4-2024-release-wrap-up)

## 📰 In the news

### ⚡ Indexing vector embeddings 7x faster

How integrating binary quantization allowed Meilisearch to be even faster.

[Read the blog post](https://blog.kerollmops.com/meilisearch-indexes-embeddings-7x-faster-with-binary-quantization)

### 🔍 Semantic search with OpenAI embeddings

Learn to use OpenAI embeddings to build semantic search.

[Read the guide](https://www.meilisearch.com/docs/guides/ai/openai)

### 🛍️ Building ecommerce search with Nuxt

Learn how to integrate real time sorting, filtering, and faceted search with Nuxt 3.

[Read the guide](https://www.meilisearch.com/blog/nuxt-ecommerce-search-guide)

### 🎓 A complete guide to predictive search

Key concepts, optimization techniques, and real-world examples to enhance UX.

[Read the blog post](https://www.meilisearch.com/blog/predictive-search)

## 🤗 Made with Meilisearch

[Fintable](https://fintable.io/) enables syncing bank transactions to spreadsheets. On their [API Coverage page](https://fintable.io/coverage), they use Meilisearch to allow searching through more than 15,000+ bank API connections.

![Searching on Fintable](https://unable-actionable-car.media.strapiapp.com/fintable_d4d4483e90.gif)

We’re always looking to promote new projects from the community. Post yours in the [#built-with-meilisearch](https://discord.meilisearch.com/) channel!


## 🤓 Did you know?

The fastest way to get started is with Meilisearch Cloud. [Start a free trial](https://meilisearch.com/cloud).

---

Stay in the loop of everything Meilisearch by subscribing to our [monthly newsletter](https://dub.sh/meili-newsletter?utm_campaign=newsletter&utm_source=blog&utm_medium=december-24-updates). To learn more about Meilisearch's future and help shape it, take a look at our [roadmap](https://roadmap.meilisearch.com/?utm_campaign=newsletter&utm_source=blog&utm_medium=december-24-updates) and participate in our [Product Discussions](https://github.com/orgs/meilisearch/discussions).

For anything else, join our developers community on [Discord](https://dub.sh/meili-discord?utm_campaign=newsletter&utm_source=blog&utm_medium=december-24-updates).

Meilisearch December updates

We're excited to introduce Meilisearch v1.12. In this article, we'll cover the most significant changes. For a full list of updates, visit the [changelog](https://github.com/meilisearch/meilisearch/releases/tag/v1.12.0) on GitHub.

<Callout>
Meilisearch 1.12 is available on Meilisearch Cloud, too—upgrade now!
</Callout>

## New: `facetSearch` and `prefixSearch` index settings

Meilisearch 1.12 introduces new APIs to disable [facet search](https://www.meilisearch.com/docs/learn/filtering_and_sorting/search_with_facet_filters) and [prefix search](https://www.meilisearch.com/docs/learn/engine/prefix). When disabled, both these settings allow Meilisearch to skip building the required data structures to enable these features. As a result, the indexing process gets much faster.

<Callout>
Disabling these features can impact search result relevancy. Be sure to assess whether your product relies on them before proceeding.
</Callout>

### Updating facet search index settings

Use the new API to [update facet search](https://www.meilisearch.com/docs/reference/api/settings#update-facet-search-settings) index settings:

```ts
// disable facet search
client.index('my_index').setFacetSearch(false)
// enable facet search (default)
client.index('my_index').setFacetSearch(true)
```

### Updating prefix search index settings

Use the new API to [update prefix search](https://www.meilisearch.com/docs/reference/api/settings#update-prefix-search-settings) index settings:

```ts
// disable prefix search
client.index('my_index').setPrefixSearch('disabled')
// enable prefix search (default)
client.index('my_index').setPrefixSearch('indexingTime')
```

## New: batches API

Meilisearch 1.12 introduces a new [batches API](https://www.meilisearch.com/docs/reference/api/batches) to retrieve data about batches. A batch is a set of [tasks](https://www.meilisearch.com/docs/learn/async/asynchronous_operations) processed together by Meilisearch.

Here is an example batch object returned (as of Meilisearch 1.12.0):

```json
{
  "uid": 160,
  "progress": {
    "steps": [
      {
        "currentStep": "processing tasks",
        "finished": 0,
        "total": 2
      },
      {
        "currentStep": "indexing",
        "finished": 2,
        "total": 3
      },
      {
        "currentStep": "extracting words",
        "finished": 3,
        "total": 13
      },
      {
        "currentStep": "document",
        "finished": 12300,
        "total": 19546
      }
    ],
    "percentage": 37.986263
  },
  "details": {
    "receivedDocuments": 19547,
    "indexedDocuments": null
  },
  "stats": {
    "totalNbTasks": 1,
    "status": {
      "processing": 1
    },
    "types": {
      "documentAdditionOrUpdate": 1
    },
    "indexUids": {
      "movies": 1
    }
  },
  "duration": null,
  "startedAt": "2024-12-12T09:44:34.124726733Z",
  "finishedAt": null
}
```

Please note that the [get batches API](https://www.meilisearch.com/docs/reference/api/batches#get-batches) accepts the same parameters as the [get tasks API](https://www.meilisearch.com/docs/reference/api/tasks#get-tasks), allowing you to retrieve batches by filtering by task data. 

For example, you can use `GET /batches?uid=123` to retrieve batches containing tasks with a `taskUid` of `123`.

### Tasks API update

For better integration, the [tasks API](https://www.meilisearch.com/docs/reference/api/tasks) now includes a `batchUid` field in task objects. You can use this value to retrieve details about a specific batch using the API to [get one batch](https://www.meilisearch.com/docs/reference/api/batches#get-one-batch).
## Indexing speed improvements 

Meilisearch 1.12 brings indexing speed improvements across the board for servers making use of multiple cores and having good I/O throughput.

This new release brings:
- 2x faster raw document insertion
- 4x faster incremental document updates in large databases
- 1.5x faster embedding generation

Additionally, this release makes task cancellation faster. On less powerful machines, indexing time remains similar but may show slight improvements.

These advancements build upon the indexing customization and speed enhancements introduced in [Meilisearch 1.6](https://www.meilisearch.com/blog/meilisearch-1-6).

## Contributors shout-out

We want to give a massive thank you to the external contributors who made this release possible. Thanks to [@airycanon](https://github.com/airycanon), [@DerTimonius](https://github.com/DerTimonius), [@flevi29](https://github.com/flevi29), [@LukasKalbertodt](https://github.com/LukasKalbertodt), [@PedroTurik](https://github.com/PedroTurik), and [@wuast94](https://github.com/wuast94) for their contribution to [Meilisearch](https://github.com/meilisearch/meilisearch), and [@dqkqd](https://github.com/dqkqd) and [@tats-u](https://github.com/tats-u) for their work on [charabia](https://github.com/meilisearch/charabia).

We also want to extend special thanks to our SDK maintainers working hard to make Meilisearch available across many [languages and frameworks](https://www.meilisearch.com/docs/learn/what_is_meilisearch/sdks).

---

And that’s a wrap for v1.12! These release notes only highlight the most significant updates. For an exhaustive listing, read the [changelog](https://github.com/meilisearch/meilisearch/releases/tag/v1.12.0) on GitHub.

Upgrade with ease using Meilisearch Cloud.

[Open Meilisearch Cloud](https://cloud.meilisearch.com/login)

---

For more information, subscribe to our [monthly newsletter](https://dub.sh/meili-newsletter?utm_campaign=release-v1-12&utm_source=blog), check out the [roadmap](https://roadmap.meilisearch.com/?utm_campaign=release-v1-12&utm_source=blog), or join our [Product Discussions](https://github.com/orgs/meilisearch/discussions).

For anything else, join our developers community on [Discord](https://discord.meilisearch.com/?utm_campaign=release-v1-12&utm_source=blog).

Meilisearch 1.12

Fuzzy search: a comprehensive guide to implementation

## 🔍 Improved Search Preview

[![Meilisearch Search Preview](https://unable-actionable-car.media.strapiapp.com/meilisearch_search_preview_v2_3b8196bc8b.gif)](https://www.meilisearch.com/docs/learn/getting_started/search_preview)

We’re improving the **Search Preview**. This new version allows further testing of your full-text search **filtering, sorting, and faceting configuration**. Plus, it now comes with a built-in **hybrid search** and **semantic ratio customization**.

[Learn more](https://www.meilisearch.com/docs/learn/getting_started/search_preview)

## ✨ Revamped index settings

[![Meilisearch index settings](https://unable-actionable-car.media.strapiapp.com/meilisearch_index_settings_v2_c5c37c4b21.gif)](https://www.meilisearch.com/docs/learn/configuration/configuring_index_settings)

Our new index settings UI allows easier, **less error-prone configuration** of your  Meilisearch settings. We want to **make Meilisearch welcoming for everyone on your team**—even for those who don't speak JSON fluently.

[Learn more](https://www.meilisearch.com/docs/learn/configuration/configuring_index_settings)

## 💫 Release week recap

[![Meilisearch Q4 2024 release wrap-up](https://unable-actionable-car.media.strapiapp.com/q4_2_24_blog_1_ee4e231bba.png)](https://www.meilisearch.com/blog/q4-2024-release-wrap-up)

Missed our release week? We get it—it happens! Sure, we’re a little sad (we worked hard to ship all this), but no worries. You can catch up on everything you missed in our latest blog post. 

[Read the blog post](https://www.meilisearch.com/blog/q4-2024-release-wrap-up)

## 📰 In the news

### 🔧 [Opt-out settings in v1.12](https://github.com/meilisearch/meilisearch/issues/4979)

New settings to opt out of prefix search and facet search to get 15% faster indexing.

### ⚡ [6.5x faster document ingestion](https://x.com/Kerollmops/status/1859635534626201993)

[Meilisearch was too slow](https://blog.kerollmops.com/meilisearch-is-too-slow) for [@kerollmops](https://x.com/Kerollmops). So we fixed it. 

### 🔌 [Support for Strapi v5](https://github.com/meilisearch/strapi-plugin-meilisearch/releases)

Our Strapi plugin now supports Strapi v5. Shout-out to our contributors [@zoran995](https://github.com/zoran995) and [@candidosales](https://github.com/candidosales) for the awesome work. 🙌

### 🛍️ [Laravel SaaS search demo + repository](https://github.com/meilisearch/saas-demo)

Get started building with Laravel using this demo of multi-index, federated search.

### 🍿 [The greatest Meilisearch ad ever](https://x.com/meilisearch/status/1852030676813378042) (0% clickbait)

It appears our Head of Product [@ledoublegui](https://x.com/ledoublegui) has multiple... talents.

## 🤗 Made with Meilisearch

The [UTMB Index](https://utmb.world/fr/utmb-index/runner-search) allows you to search for any runner who participated in a trail running event. The site uses Meilisearch to power its instant search (with a short debounce!) Thanks to [@CartGauthier](https://x.com/CartGauthier/status/1852025207839449191) for sharing this with us.

[![UTMB index runner search](https://unable-actionable-car.media.strapiapp.com/utmb_runner_search_9b060121f2.gif)](https://x.com/CartGauthier/status/1852025207839449191)

_We’re always looking to promote new projects from the community. Post yours in the [#built-with-meilisearch](https://dub.sh/meili-discord?utm_campaign=newsletter&utm_source=blog&utm_medium=november-24-updates) channel!_

## 🦸 We're hiring

See our open positions:

- [Chief of Staff](https://www.meilisearch.com/careers)

## 🤓 Did you know?

The fastest way to get started is with Meilisearch Cloud. [Start a free trial.](https://meilisearch.com/cloud)

---

Stay in the loop of everything Meilisearch by subscribing to our [monthly newsletter](https://dub.sh/meili-newsletter?utm_campaign=newsletter&utm_source=blog&utm_medium=november-24-updates). To learn more about Meilisearch's future and help shape it, take a look at our [roadmap](https://roadmap.meilisearch.com/?utm_campaign=newsletter&utm_source=blog&utm_medium=november-24-updates) and participate in our [Product Discussions](https://github.com/orgs/meilisearch/discussions).

For anything else, join our developers community on [Discord](https://dub.sh/meili-discord?utm_campaign=newsletter&utm_source=blog&utm_medium=november-24-updates).

Meilisearch November updates


Have you ever noticed how Google suggests exactly what you're looking for before you finish typing? That's predictive search in action. It revolutionizes user interaction with software applications. In this guide, we'll explore implementing predictive search in your applications.

## Introduction to predictive search

### Definition and importance of predictive search

**Predictive search is an intelligent system that anticipates user intent and suggests relevant results in real-time as users type**. Unlike traditional search methods that wait for a complete query, predictive search processes each keystroke to provide immediate, relevant suggestions.

The importance of predictive search in software engineering is significant. It bridges user intent and content discovery, reducing the time and effort users spend searching. Modern applications leverage predictive search to enhance productivity and streamline workflows.

Software engineering teams implementing predictive search can expect improved user engagement metrics. For instance, e-commerce platforms often see a 25-30% increase in conversion rates with predictive search features. Developer tools report significant reductions in time spent searching for code snippets or documentation.

### Predictive search enhancing autocomplete

Traditional autocomplete completes words based on character matching. Predictive search, however, analyzes patterns in user behavior, search history, and contextual data to offer intelligent suggestions.

The personalization aspect of predictive search sets it apart from basic autocomplete. **Predictive search creates a more tailored experience by considering factors like previous searches, user location, and trending queries**.

For example, when a developer searches for "deploy" in a cloud platform's console, the system prioritizes suggestions based on their commonly used services or recent deployments.

<Callout intent="note">
	
	For example, in an e-commerce platform, **autocomplete might suggest "laptop bag"** when a user types "lap", predictive search might also suggest "computer accessories," "mobile workstation," or specific laptop models based on browsing history and market trends.
	
	
</Callout>

## Core concepts and components

### Predictive models in software engineering

Predictive models form the backbone of modern search capabilities. **They utilize complex algorithms to analyze user input patterns and generate relevant suggestions**.

Businesses can leverage large datasets, with the specific type of data varying depending on the industry and business objectives. It is possible to use various data sources like browsing history, user interactions, and market trends to improve predictive search strategies.

![Examples Of Data Predictive Search](https://unable-actionable-car.media.strapiapp.com/predictive_search_b4778b9493.avif)

These models process queries through multiple layers of analysis, including string matching, semantic understanding, and historical pattern recognition.

Predictive models often rely on prefix trees (or radix trees) for efficient string matching and retrieval. This structure enables rapid processing of partial queries, allowing the system to generate suggestions with minimal latency. The models continuously learn from user interactions, refining their prediction accuracy over time.

### Search engines and search queries

Search engines in predictive systems employ sophisticated indexing mechanisms to aid rapid query processing. Key components include query processors, suggestion generators, and ranking algorithms.

These elements work together to produce and prioritize results based on various relevance factors.

![Layers Of Logic Search Engine](https://unable-actionable-car.media.strapiapp.com/predictive_search_2c856b34f6.avif)

Modern search queries go through steps including tokenization, spell checking, and context analysis. **This ensures that even imperfect queries can yield useful suggestions**. Balancing immediate results and accuracy often involves using caching mechanisms to optimize frequently requested queries.

### Machine learning and AI power

Artificial intelligence significantly enhances predictive search through advanced natural language processing and machine learning algorithms.

These technologies help systems understand user intent beyond literal string matching, incorporating semantic understanding and contextual relevance.

AI improves search accuracy by:

- Capturing user intent with natural language understanding
- Adapting ranking algorithms to changing user behavior patterns
- Identifying relationships between different terms through semantic analysis
- Tailoring suggestions based on user profiles and historical interactions

Combining machine learning with traditional search algorithms creates a hybrid approach that delivers both speed and accuracy.

## Capabilities required for predictive search

A robust predictive search implementation requires several key capabilities to deliver a seamless user experience. Fast indexing and efficient query processing are foundational, but there's more to consider.

### Meilisearch's approach to enabling predictive search

Meilisearch combines traditional full-text search with AI-powered capabilities. Its architecture enables fast, relevant results and maintains simplicity in implementation.

**The integration process focuses on three key areas**: data indexing, search configuration, and result optimization. Here's an example of configuring search settings:

```typescript
// config/search.ts
const searchConfig = {
  settings: {
    rankingRules: [
      'words',
      'typo',
      'proximity',
      'attribute',
      'sort',
      'exactness'
    ],
    searchableAttributes: [
      'title',
      'description',
      'category'
    ],
    displayedAttributes: [
      'title',
      'description',
      'category',
      'thumbnail'
    ],
    // Enable AI-powered search with OpenAI
    embedder: {
      type: 'openai',
      documentTemplate: "Product: {{title}}nDescription: {{description}}"
    }
  }
};

await searchClient.index('products').updateSettings(searchConfig);
```

**This configuration enables both traditional search capabilities and AI-powered features**, creating a hybrid search experience that delivers both speed and relevance. The system handles typos, synonyms, and contextual understanding while maintaining sub-50ms response times.

<Callout intent="tip">
	
	Ready to supercharge your search experience? [Explore Meilisearch Cloud](https://meilisearch.com/cloud) and deliver lightning-fast search results that will keep your users engaged and boost your conversion rates.
	
	
</Callout>

## Applications and use cases

Modern software applications leverage predictive search in numerous ways, from e-commerce platforms to content management systems. Understanding these practical applications helps developers implement the right search strategies for their specific use cases.

### App search

**App search is a common implementation of predictive search in software applications**. Mobile apps benefit particularly due to limited screen space and typing constraints.

For instance, food delivery apps use predictive search to help users find restaurants or dishes quickly, considering factors like previous orders and current location.

### Site and media search

**Site and media search capabilities have evolved to handle complex content structures**. Content management systems and documentation platforms use predictive search to help users navigate large information hierarchies.

For example, technical documentation sites can suggest relevant API endpoints or code examples as developers type their queries.

<Callout intent="note">
	
	Check out how Meilisearch powers the discovery of 300,000+ AI models on [Hugging Face](https://www.meilisearch.com/blog/hugging-face-democratizes-ai), demonstrating the power of predictive search in organizing and accessing vast AI resources.
	
	
</Callout>

### Ecommerce search

E-commerce platforms showcase the potential of predictive search through sophisticated product discovery. Online stores suggest not only product names but also categories, brands, and specific features. **This capability extends to handling complex queries like "red summer dress under $50" accurately**.

<Callout intent="note">
	
	For another compelling example of predictive search in action, explore [Meilisearch's case study with Bookshop.org](https://www.meilisearch.com/blog/bookshop-increases-search-based-purchases), where they achieved a remarkable 43% increase in search-based purchases by enhancing their book discovery platform.
	
	
</Callout>

### Enterprise search

[Enterprise search](meilisearch.com/blog/enterprise-search) presents unique challenges because it handles diverse data sources while maintaining security constraints. Predictive search in enterprise environments must consider user permissions, department-specific terminology, and internal document hierarchies.

For instance, a company's internal tools might suggest different results for the same query based on the user's role and access level.

## Techniques and methodologies

### Genetic programming and test models

Genetic programming enhances predictive search capabilities by **evolving and refining search algorithms**. These techniques mimic natural selection to optimize search patterns and improve relevancy.

The system dynamically generates search algorithm variations. It then evaluates these algorithms by analyzing user interactions. Based on performance metrics, the system selects the most effective algorithm combinations to enhance future search experiences.

Test models developed through genetic programming adapt to changing search patterns. For instance, e-commerce search systems can automatically adjust prediction weights based on seasonal trends, user behavior shifts, and new product introductions. **The evolutionary nature of these algorithms ensures relevant search suggestions** as user preferences and patterns evolve.

### Base software and search base

Establishing a robust search base requires careful consideration of data structure and indexing strategies.

<Callout intent="note">
	
	For deeper insights into optimizing search indexing, check our comprehensive guide on [best practices for faster indexing](https://meilisearch.com/blog/best-practices-for-faster-indexing).
	
	
</Callout>

Typically, it consists of an inverted index system enabling quick lookup of terms while maintaining relationships between search elements. **This base structure must support both exact matches and fuzzy searching capabilities**.

Integration of predictive search within software frameworks demands:

- Efficient data storage and retrieval mechanisms to handle large volumes of search queries
- Flexible indexing strategies accommodating both structured and unstructured data
- Scalable architecture maintaining performance under increasing load
- Real-time update capabilities to reflect changes in underlying data

The search base must also handle edge cases and special scenarios. For example, with multilingual search capabilities, the base system must account for different character sets, writing systems, and language-specific patterns. **This ensures reliable and effective predictive search functionality across diverse use cases** and user requirements.

## Capabilities

Low latency is crucial for predictive search systems because users expect instant feedback while typing. Modern search implementations must maintain response times under 100ms to feel instantaneous. This speed requirement becomes even more critical when handling complex queries across large datasets.

Several optimization techniques help achieve these performance goals. **Intelligent caching strategies store frequently accessed results and partial queries**, significantly reducing database load and response times for common searches. Here's how response times typically break down:

| Operation | Target Latency | Optimization Method |
| --- | --- | --- |
| Query Processing | &lt; 20ms | In-memory processing |
| Index Lookup | &lt; 30ms | Cached indices |
| Result Ranking | &lt; 30ms | Pre-computed scores |
| Network Transfer | &lt; 20ms | Edge caching |

Ensuring reliability involves implementing robust error handling and failover mechanisms. **Search systems must gracefully handle scenarios like network issues, invalid queries, or backend service disruptions**. Implementing circuit breakers and fallback options helps maintain service availability.

Performance monitoring is vital for maintaining optimal search functionality. Modern search implementations use real-time metrics tracking to identify bottlenecks and optimization opportunities.

Key performance benchmarks for predictive search systems include:

- **Query response time:** 95th percentile under 200ms
- **Suggestion relevancy:** Minimum 80% accuracy rate for top 5 suggestions
- **System throughput:** Ability to handle 1000+ queries per second per node
- **Cache hit ratio:** At least 85% cache effectiveness

<Callout intent="tip">
	
	Meilisearch Cloud includes [built-in monitoring tools](https://www.meilisearch.com/docs/learn/analytics/monitoring) that enable real-time tracking of search performance metrics.
	
	
</Callout>

The architecture should scale horizontally to handle increasing load while maintaining performance. This typically involves:

1. Distributed index sharding for better resource utilization
2. Load balancing across multiple search nodes
3. Asynchronous index updates to prevent blocking operations
4. Smart query routing based on user location and server load

Reliability also extends to data consistency and accuracy. Search systems must handle index updates and real-time data changes without compromising search quality or availability. **This often requires sophisticated versioning and synchronization mechanisms** to ensure users always see the most relevant and up-to-date results.

## Future directions and innovations

As predictive search technology continues to evolve, several emerging trends and innovations are reshaping the future of search systems.

### Integration with advanced technologies

The future of predictive search is being shaped by **advancements in artificial intelligence and machine learning technologies**. Large language models (LLMs) are transforming how search systems understand and process user queries. These models can interpret natural language with high accuracy, enabling more conversational and context-aware search experiences.

Integration with advanced technologies is already showing promising results. Some systems are experimenting with multimodal search capabilities, processing text, images, and voice inputs simultaneously. This enables more intuitive search interfaces that match how humans naturally communicate.

### Trends in predictive search

LLMs are revolutionizing predictive search capabilities by fundamentally transforming how we interact with search engines. These sophisticated systems can now comprehend complex queries and generate natural, contextually appropriate suggestions that feel remarkably human-like.

Modern search systems now interpret both technical and informal language, moving beyond pattern matching to understand the true meaning behind user queries.

This advancement has enabled several crucial improvements in search functionality:

- More precise interpretation of user queries
- Sophisticated handling of ambiguous searches
- Enhanced recognition of specialized terminology
- Intelligent, context-aware result ranking

**These innovations are already being implemented in various forms**. Some search engines use hybrid approaches that combine traditional search algorithms with [AI-powered semantic understanding](meilisearch.com/blog/ai-powered-search). This allows them to maintain the speed and reliability of conventional search while leveraging advanced AI capabilities.

## Beyond horizons: the transformative potential of predictive search

Predictive search has evolved from a convenient feature to an essential component of modern software applications.

Successful implementation requires careful consideration of architecture, performance optimization, and user experience design.

The integration of advanced AI technologies and mobile-first approaches will continue to push the boundaries of what's possible in predictive search. This makes it an exciting area for continued innovation and development.

<Callout intent="tip">
	
	Unlock advanced search potential with customizable relevancy, typo tolerance, and more. [Enhance Your Search Strategy](https://meilisearch.com/features) with Meilisearch Cloud's powerful search capabilities.
	
	
</Callout>

Software engineering predictive search: a complete guide

_This blog post was originally published on kero's [personal blog](https://blog.kerollmops.com/meilisearch-indexes-embeddings-7x-faster-with-binary-quantization)._

<Callout intent="note">
This is one blog post in a series:
- [Spotify-Inspired: Elevating Meilisearch with Hybrid Search and Rust](https://blog.kerollmops.com/spotify-inspired-elevating-meilisearch-with-hybrid-search-and-rust)
- [Multithreading and Memory-Mapping: Refining ANN Performance with Arroy](https://blog.kerollmops.com/multithreading-and-memory-mapping-refining-ann-performance-with-arroy-in-rust)
- [Meilisearch Expands Search Power with Arroy's Filtered Disk ANN](https://blog.kerollmops.com/meilisearch-expands-search-power-with-arroy-s-filtered-disk-ann)
- [How Meilisearch Updates a Millions Vector Embeddings Database in Under a Minute](https://blog.kerollmops.com/how-meilisearch-updates-a-millions-vector-embeddings-database-in-under-a-minute)
- Meilisearch Indexes Embeddings 7x Faster with Binary Quantization.
</Callout>

For the past year at Meilisearch, [we’ve been working hard on hybrid search](https://github.com/meilisearch/meilisearch/issues/3838), which mixes keyword search with semantic search using [Arroy: our vector store, based on a Spotify technology](https://blog.kerollmops.com/spotify-inspired-elevating-meilisearch-with-hybrid-search-and-rust). A vector store is a data structure that efficiently stores embeddings (vectors) for fast and relevant retrieval, called Approximate Nearest Neighbors (ANN) search.

When our customers started to use arroy heavily, some of them started reaching the machine limits. For example, for a customer with a 768 dimensions model, we found out that we couldn’t index more than 15M of embeddings on a machine with 64GiB of RAM.

### Binary Quantization to the Rescue

The concept behind this elaborate term is that we are going to *quantize* the embeddings. Quantizing a number involves spreading it into a defined number of values; for *binary* quantization, two values can be represented on a single bit, e.g., 0.2561 becomes 1, and -0.568 becomes -1.

This implies that we could convert a 32-bit float number into a 1-bit number, which would divide the disk and RAM usage by 32. That would mean we could index up to 480M embeddings with 64GiB of RAM, instead of being limited to 15M embeddings.

An issue we faced is that bits cannot represent -1 and 1 but only 0 and 1. We had to use some tricks to make it work and fake 0s into -1s in our computations. For example, one of the functions that we had to optimize a lot was the function in charge of converting a binary quantized vector to 32-bit float vectors.

We do not publicly expose raw quantized embeddings and utilize this function to provide real 32-bit float embeddings to the users. However, the primary reason for converting 1-bit quantized embeddings to 32-bit floats is to ensure high precision when [generating hyperplanes to group embeddings together](https://blog.kerollmops.com/spotify-inspired-elevating-meilisearch-with-hybrid-search-and-rust#spotify-s-hyperplane-trees-for-efficient-anns) and maintaining relevant ANNs.

![Profiling backtrace showing slow 1-bit to 32-bit float conversion](https://unable-actionable-car.media.strapiapp.com/68747470733a2f2f692e696d6775722e636f6d2f497165713649562e706e67_596adc5fd5.png)

_Converting a quantized embedding back to 32-bit floating points takes up to 24.21s on a 1m53s run._

Below is the SIMD neon-specific version of this function. It processes raw bytes corresponding to a 1-bit quantized embedding, converting them into a properly aligned 32-bit float embedding. Using masks, the function separates the low and high bits, processing them in batches of four in 128-bit registers, which are then transformed into four 32-bit floats. These values are collectively written to the output vector.

```rust
unsafe fn to_vec_neon(bytes: &[u8]) -> Vec<f32> {
    let mut output: Vec<f32> = vec![0.0; bytes.len() / u32::BITS];
    let output_ptr = output.as_mut_ptr();
    let low_mask = [0b_0000_0001, 0b_0000_0010, 0b_0000_0100, 0b_0000_1000];
    let high_mask = [0b_0001_0000, 0b_0010_0000, 0b_0100_0000, 0b_1000_0000];
    let ones = unsafe { vld1q_dup_f32(&1.0) };
    let minus = unsafe { vld1q_dup_f32(&-1.0) };

    for (current_byte, base) in bytes.iter().enumerate() {
        unsafe {
            let base = *base as u32;
            let base = vld1q_dup_u32(&base);
            for (i, mask) in [low_mask, high_mask].iter().enumerate() {
                let mask = vld1q_u32(mask.as_ptr());
                let mask = vandq_u32(base, mask);
                // 0xffffffff if equal to zero and 0x00000000 otherwise
                let mask = vceqzq_u32(mask);
                let lane = vbslq_f32(mask, minus, ones);
                let offset = output_ptr.add(current_byte * 8 + i * 4);
                vst1q_f32(offset, lane);
            }
        }
    }

    output
}
```

_<a href="https://x.com/lemire/status/1835700368648323261">Daniel Lemire doesn't seem to have found a better algorithm</a>_ 🤭

### The Results

We chatted about the theoretical benefits of this approach earlier, but how does it work out in real life? Here are different tables summarizing the results for you.

<details>
  <summary>Open to see the original benchmarks</summary>
  <img src="https://unable-actionable-car.media.strapiapp.com/68747470733a2f2f692e696d6775722e636f6d2f67667765626b592e706e67_a96a995c7c.png"  alt="Original arroy benchmark results"/>
</details>

#### Embeddings with 1024 Dimensions

When examining small embeddings, we can observe a significant impact on relevancy due to binary quantization. The primary advantages of this approach lie in faster insertion times and reduced storage requirements. It is not advisable to enable this feature for small embeddings, as they already occupy minimal disk space. Storing over four million 1024-dimensional embeddings requires just 8 GiB of disk space.

| Version                 | Recall @1 | Recall @20 | Recall @100 | Indexing time | Search time | On disk size |
|-------------------------|-----------|------------|-------------|---------------|-------------|--------------|
| Cosine                  | **1.00**  | **0.72**   | **0.84**    | 491s          | **23s**     | 8 GiB        |
| Binary Quantized Cosine | 0.87      | 0.50       | 0.52        | **147s**      | 27s         | **850 MiB**  |

#### Embeddings with 1536 Dimensions

Binary quantization proves advantageous for managing larger embeddings. It significantly reduces search times, utilizing just 15% of the original disk space. Indexing about 500,000 1536-dimensional embeddings is achieved in a quarter of the time. Despite this, there remains a noticeable impact on relevancy...

| Version                 | Recall @1 | Recall @20 | Recall @100 | Indexing time | Search time | On disk size |
|-------------------------|-----------|------------|-------------|---------------|-------------|--------------|
| Cosine                  | **1.00**  | **0.82**   | **0.90**    | 1219s         | 50s         | 13 GiB       |
| Binary Quantized Cosine | **1.00**  | 0.70       | 0.70        | **235s**      | **47s**     | **2 GiB**    |

#### Embeddings with 3072 Dimensions

The impact of binary quantization on large embeddings is evident. While relevance may be slightly affected, the reduction in disk usage, search time, and indexing time is significant. Being able to store the same number of embeddings with just 15% of the space and index 6 times faster is a major achievement! It takes half as much disk space to store a million 3072-dimensional binary quantized embeddings than four million 1024-dimensional embeddings.

| Version                 | Recall @1 | Recall @20 | Recall @100 | Indexing time | Search time | On disk size |
|-------------------------|-----------|------------|-------------|---------------|-------------|--------------|
| Cosine                  | **1.00**  | **0.86**   | **0.92**    | 4000s         | 150s        | 26 GiB       |
| Binary Quantized Cosine | **1.00**  | 0.77       | 0.77        | **560s**      | **100s**    | **4 GiB**    |

### Conclusion

The impact of binary quantization on recall was expected to decrease, given the reduced information in the algorithm. However, for datasets with over 1500 dimensions, the impact on recall is relatively minor compared to the significant benefits in disk space and indexing speed.

In practice, we observed that space utilization decreased not by 32 times but by just 15% of the original space, a notable achievement. This is primarily due to arroy storing not only user vectors but also internal nodes for efficient searching. Split nodes (hyperplanes) are retained as full 32-bit floats to maintain precision.

- Indexing speed improves nearly 5 times with 1536 dimensions and about 7 times with 3072 dimensions.
- Disk size shrinks to around 6.5 times smaller.
- Search time shows slight improvements for embeddings starting at 1536 dimensions.

We recommend using Meilisearch/arroy's binary quantization feature for OpenAI users with large embeddings. In an upcoming article, we will compare Meilisearch/arroy with competitors, focusing on our Filtered DiskANN feature. Stay tuned!

Meilisearch indexes embeddings 7x faster with binary quantization

We've been busy working on things you've asked for, and the team is excited to share what's new in Meilisearch this quarter.

## A revamped search preview 🔍

The [search preview](https://www.meilisearch.com/docs/learn/getting_started/search_preview) has been one of our most-used features, but it didn't let you test everything Meilisearch can do. We've now rebuilt our search preview to make it more useful to assess test search result relevancy when you are tweaking an index's settings.

What's included:

- Show/hide the results' [ranking score](https://www.meilisearch.com/docs/learn/relevancy/ranking_score):

![ranking score.png](https://unable-actionable-car.media.strapiapp.com/image_38_9583f10fe8.png)

- Check and customize search options
- Configure filterable attributes as facets
- Test [AI-powered searches](https://www.meilisearch.com/docs/learn/ai_powered_search/getting_started_with_ai_search) if this functionality has been enabled for your project.

An updated preview allows you to test everything in real-time, making it easier to get your search working exactly how you want.

## New index settings in the UI ⚙️

We've fixed one of your biggest frustrations - you and your team no longer need to rely on JSON input to [change your index settings](https://www.meilisearch.com/docs/learn/configuration/configuring_index_settings). Instead of typing JSON code, you'll now see straightforward controls in the UI for adjusting settings like:

- Add and delete searchable attributes:
![searchable attributes/index settings.png](https://unable-actionable-car.media.strapiapp.com/image_37_5bea152778.png)

- Change the ranks of your searchable attributes
- Reset to default

Now anyone on your team can get access to user-friendly modification of the index settings without needing to understand JSON syntax.

## Speed up your search  with feature opt-out 🚀

We've got some updates for those of you working with massive datasets. To **reduce disk usage** and **speed up indexing by up to 15%**, you can now optimize Meilisearch's performance: 

- Choose to opt-out of prefix search and facet search features
- Reduce indexing time and save disk space

This has been one of our most requested features from users dealing with large-scale deployments, and we're excited to see how you'll use it.

## What else is cooking? 🎯
#### New Japan region is coming in Q4! 🇯🇵
Soon, you’ll be able to host your Meilisearch projects in Japan. This means much faster searches for users in Asia-Pacific - instead of connecting to Singapore, they'll connect directly to servers in Japan. The difference? A search that used to take 80+ milliseconds will happen almost instantly.

To use it, you’ll just need to pick Japan when you create a new project in the [cloud](https://cloud.meilisearch.com) dashboard. Keep an eye on our updates to know exactly when it’s released!


We are not stopping here. Our team is already working on our early 2025 releases. Here's a quick spoiler:

- AI-powered hybrid search general availability
- New Cloud dashboard features
- Even better performance optimizations

## Try it out!

Want to test these features? [Sign up for Meilisearch Cloud](https://cloud.meilisearch.com/) to experience these new features firsthand.

---

We built these changes based on your feedback, and we'd love to know if they help. Try them out and tell us what you think.

For  a more in-depth look at these updates, visit our [documentation](https://docs.meilisearch.com/).

Meilisearch Q4 2024 release wrap-up

##  🐿️ Meilisearch 1.11 release

![Release notes - Meilisearch 1.11](https://unable-actionable-car.media.strapiapp.com/meilisearch_1_11_59ea013828.jpg)

Meilisearch 1.11 introduces key **improvements** in **AI-powered search**, including **binary quantization**. This release also **enhances federated search** functionality based on user feedback and introduces a **new filter operator**.

[Discover updates](https://www.meilisearch.com/blog/meilisearch-1-11)

***


## 🌎 TutKit's journey with Meilisearch 

[![Meilisearch powers multilingual learning at scale for Tutkit](https://unable-actionable-car.media.strapiapp.com/Tut_Kit_516315e429.png)](https://www.meilisearch.com/blog/tutkit-multilingual-learning-at-scale)

Discover how Tutkit.com scaled its learning platform to handle search across **15,000+ resources in 26 languages**.

[Read case study](https://www.meilisearch.com/blog/tutkit-multilingual-learning-at-scale)

***

## 🎥 Webinar recording
[![October 9th.png](https://unable-actionable-car.media.strapiapp.com/October_9th_b09a084f13.png)](https://www.youtube.com/watch?v=CvutT64u0fE)


Missed our October event? No worries—we've got you covered with the full recording!

 

Discover Meilisearch's **roadmap** and latest **updates**, learn more about **federated search and its benefits** and see a live demo of the **new Cloud functionality**.

[Watch it now!](https://www.youtube.com/watch?v=CvutT64u0fE)

***

## 📰 In the news
### ⚛️  [Add AI-powered search to React](https://www.meilisearch.com/blog/add-ai-powered-search-to-react)
Build a React movie search and recommendation app with Meilisearch's AI-powered search.

### 🔍 [Federated search guide](https://www.meilisearch.com/docs/learn/multi_search/performing_federated_search)

Learn how to create separate indexes for different CRM data types and query them simultaneously for a unified results list.

### ✨ [Semantic search with Cloudflare embeddings](https://www.meilisearch.com/docs/guides/ai/cloudflare)

Learn how to set up Meilisearch with Cloudflare Worker AI embeddings to enable semantic search capabilities.

### 🛍️ [Shopify storefront guide for Next.js](https://www.meilisearch.com/blog/next-js-shopify-storefront-with-blazity)

Learn to build a Shopify storefront with Next.js and the [Blazity commerce starter](https://github.com/Blazity/enterprise-commerce), using Meilisearch hybrid search to serve all front-end data.

### 🔒 [Laravel multitenancy guide](https://www.meilisearch.com/docs/guides/security/laravel_multitenancy)

How to implement search in a multitenant Laravel application.

***

## 🤗 Made with Meilisearch

A community member built [Firebits.io](https://firebits.io) and shared it with the Discord community. Firebits is a modern online Second Brain that helps you organize and share your online learning resources. Its search is, of course, powered by Meilisearch 🎉 

![firebits-2.gif](https://unable-actionable-car.media.strapiapp.com/firebits_2_869f9caaa0.gif)

We’re always looking to promote new projects from the community. Post yours in the [#built-with-meilisearch](https://discord.meilisearch.com/) channel!

***

## 🦸‍♀️ We’re hiring!

See our open positions:

- [Chief of Staff](https://jobs.lever.co/meili/4abb7650-a5ca-4aac-9f35-e2fe5144b78d)

***

## 🤓 Did you know?

The fastest way to try Meilisearch is our [Cloud free trial](https://www.meilisearch.com/cloud).

[Get started!](https://www.meilisearch.com/cloud)

***

Stay in the loop of everything Meilisearch by subscribing to our [monthly newsletter](https://dub.sh/meili-newsletter?utm_campaign=newsletter&utm_source=blog&utm_medium=october-24-updates). To learn more about Meilisearch's future and help shape it, take a look at our [roadmap](https://roadmap.meilisearch.com/) and participate in our [Product Discussions](https://github.com/orgs/meilisearch/discussions).

For anything else, join our developers community on [Discord](https://discord.meilisearch.com/).



Meilisearch latest news and company updates