Marvin - Detailed Review
Data Tools
Marvin - Product Overview
Marvin Overview
Marvin is a lightweight AI toolkit that simplifies the integration of natural language interfaces into various software projects.
Primary Function
Marvin’s primary function is to enable developers to build reliable, scalable, and trustworthy natural language interfaces. It does this by automating the interaction between software and large language models (LLMs), allowing developers to focus on writing clean, versioned, and reusable code rather than crafting complex prompts.
Target Audience
The target audience for Marvin is developers who want to incorporate AI capabilities into their software projects without the need to build AI from scratch. This includes engineers who are familiar with Python and want to leverage LLMs for tasks such as text classification, entity extraction, and data generation.
Key Features
- Simplicity and Self-Documentation: Each tool in Marvin is simple and self-documenting, making it easy for developers to use without extensive AI expertise.
- Multi-Modal Support: Marvin supports both text and image inputs, as well as audio and image generation, providing flexibility in how data can be processed and generated.
- Independent Tools: The tools within Marvin are independent and can be used individually or in combination with other libraries, allowing for incremental adoption.
- Automated Translation: Marvin handles the translation between code and LLM prompts, so developers can write code as they normally would without worrying about crafting the right prompts.
- Custom AI-Powered Behaviors: Developers can combine inputs, instructions, and output types to create custom AI behaviors, such as classifying text, extracting product features from user feedback, and generating synthetic data.
Conclusion
Overall, Marvin is a practical tool for developers who want to integrate AI capabilities into their projects efficiently and reliably.
Marvin - User Interface and Experience
User Interface
Marvin’s interface is centered around natural language interactions, making it accessible and user-friendly. Here, the traditional front-end and back-end architecture is redefined by using Large Language Models (LLMs) as the primary interface. This approach allows users to interact with the system through conversational inputs, eliminating the need for a traditional API.
For example, in applications like the “BookMate” or “ToDo App,” users engage with the system by providing natural language instructions. The LLM then interprets these instructions, manages the application’s state, and provides relevant responses or updates.
Ease of Use
Marvin is engineered to be easy to use, especially for developers who are familiar with Python. It allows users to write code as they normally would, without the need to craft specific prompts for LLMs. Marvin handles the translation between the code and the LLM, making it simple to integrate AI capabilities into traditional software projects.
The tools within Marvin are self-documenting and independent, allowing users to adopt them incrementally. This flexibility ensures that users can focus on writing clean, versioned, and reusable code, rather than worrying about the intricacies of AI interactions.
User Experience
The overall user experience with Marvin is streamlined and efficient. The system is multi-modal, supporting both text and other media types like images and audio, which enhances its versatility.
In the context of user research, Marvin’s “Ask AI” feature provides a powerful search function that allows users to ask specific questions about their research data. This feature can summarize insights, detect themes and trends, and even compare different topics, all within a user-friendly interface.
Marvin also ensures data security and privacy, which is crucial for maintaining trust and confidence in the system. By centralizing user feedback and research data, Marvin helps users make better decisions based on accurate and relevant insights.
In summary, Marvin’s user interface is characterized by its natural language-based interactions, ease of use through straightforward coding, and a comprehensive user experience that simplifies the integration of AI into various applications.
Marvin - Key Features and Functionality
Marvin: A Python Library for Integrating Large Language Models
Marvin is a Python library that integrates Large Language Models (LLMs) into various data tools and functions, offering several key features that simplify working with AI. Here are the main features and how they work:AI Functions
Marvin’s AI functions allow you to create custom functions without writing any source code. These functions leverage LLMs to predict outputs based on the function’s name, description, signature, and inputs.Example
You can create a function to analyze sentiment by decorating a Python function with `@marvin.fn`. For instance: “`python @marvin.fn def sentiment(text: str) -> float: “””Returns a sentiment score for `text` between -1 (negative) and 1 (positive).””” “` When called, this function uses the LLM to generate the sentiment score without executing any source code.Functionality and Benefits
- Versatility: AI functions can handle complex tasks involving natural language processing, such as sentiment analysis, recipe generation, and more.
- Safety: Since no source code is generated or executed, this approach ensures safety for a wide range of use cases.
- Ease of Use: Users only need to define the function’s form, including its name, description, and type hints, and the AI takes care of the rest.
Assistants
Marvin’s assistants API allows for interactive and stateful conversations with LLMs.Conversational Interaction
Assistants can maintain context, state, and multiple threads of conversation, making them more powerful than traditional request/response chat APIs.Custom Tools
You can provide custom tools to assistants, such as functions that fetch web pages or perform other tasks. This is done by passing Python functions to the assistant’s `tools` argument. “`python from marvin.beta.assistants import Assistant ai = Assistant(name=”Marvin”, instructions=”You the Paranoid Android.”) ai.say(‘Hello, Marvin!’) “`Lifecycle Management
Assistants can be managed with different lifecycle strategies, such as lazy or context-based management, which affect how the assistant is registered with the OpenAI API.High-Level Components
Marvin provides several high-level components to simplify working with AI:Model Parsing
Marvin can turn your function into a prompt, use AI to get its most likely output, and parse the response. For example, you can use Marvin to parse a location from a text description: “`python import marvin from pydantic import BaseModel class Location(BaseModel): city: str state: str latitude: float longitude: float marvin.model(Location)(“They say they’re from the Windy City!”) “`Classifiers
You can define classifiers that categorize text into specific categories without writing any code. “`python from marvin import classifier from typing import Literal @classifier def customer_intent(text: str) -> Literal: “””Classifies incoming customer intent””” customer_intent(“I need to pick up my prescription”) # “Pharmacy” “`List Generators
Marvin can generate lists based on the function’s description and inputs. “`python @marvin.fn def list_fruits(n: int, color: str = ‘red’) -> list: “””Generates a list of fruits””” list_fruits(3) # “” “`Benefits
- Ease of Integration: Marvin integrates seamlessly into Python code, making it easy to use AI without needing to write complex code.
- Flexibility: The library supports a wide range of tasks, from simple data transformations to complex natural language processing.
- Safety and Efficiency: By not generating or executing source code, Marvin ensures a safe and efficient way to leverage AI for various tasks.
Marvin - Performance and Accuracy
Performance
Marvin demonstrates impressive performance in several areas:Efficient Document Retrieval
Marvin uses vector embeddings and multimodal AI to enable more accurate and efficient document retrieval based on semantic similarity. This allows users, such as bank employees, to quickly retrieve relevant documents without needing exact keyword matches.Time Savings
The platform significantly reduces the time spent on analyzing data. Users spend 60% less time analyzing research data, thanks to features like automatic note-taking, AI-powered search, and advanced transcription models.Advanced Content Classification
Marvin accurately categorizes documents and other data types based on their content and context, which is particularly beneficial for industries like insurance where claims documents need to be classified quickly.Enhanced Summarization
The AI generates insightful and comprehensive document summaries, helping users like customer service representatives to quickly get accurate answers to customer queries.Accuracy
Marvin’s accuracy is enhanced by several advanced AI models:Semantic Search
The platform supports advanced semantic search, allowing users to retrieve information based on meaning and context rather than just keywords.High-Quality Transcriptions
Marvin integrates state-of-the-art speech recognition models, such as AssemblyAI’s Conformer-2, which achieve near human-level performance in transcribing audio and video data.Data Analysis
The AI models embedded in Marvin help in collecting, organizing, analyzing, and sharing qualitative research data accurately, which is crucial for building customer-centric products.Limitations and Areas for Improvement
Despite its strengths, Marvin has some limitations and areas that need improvement:Learning Curve
Marvin has a specific way of operating, which can be confusing for users. The tagging taxonomy, for example, can be unclear, and the platform’s complexity may require significant time to learn.Context and Empathy
While Marvin excels in pattern recognition, it lacks the deep understanding of company context, industry knowledge, and user empathy that human researchers possess. It also struggles to understand the underlying motivations and reasons behind user behavior.Potential for Errors
Marvin’s AI can make mistakes or “hallucinate” information, especially with ambiguous or complex data. This highlights the need for human oversight and validation.Data Security Concerns
Using AI tools like Marvin raises questions about data privacy and security, particularly when dealing with sensitive user information. However, Marvin does include features like PII redaction to address these concerns.Pricing and Accessibility
Marvin’s pricing tiers are not friendly to startups or small businesses, as they require a minimum of 5 users for paid plans. This can be a significant barrier for smaller organizations. In summary, Marvin offers significant improvements in efficiency and accuracy for data analysis and document retrieval, but it also comes with a steep learning curve and some limitations in terms of context understanding and error potential. Addressing these areas could further enhance its usability and reliability. Marvin - Pricing and Plans
The Pricing Structure of Marvin
Marvin, an AI-driven tool for user research, is segmented into several plans, each with distinct features and limitations.
Free Plan
- This plan is free to use forever and requires no credit card.
- It includes 5 files per month, 2 contributors, and unlimited viewers.
- Call recordings are limited to 40 minutes per session.
Essentials Plan
- This plan starts at $50 per user per month, with a minimum of 5 users.
- It offers 30 files per month, 5 contributors, and 5 viewers.
- This plan is suitable for startups looking to build a culture of design and research.
Standard Plan
- Priced at $100 per user per month, with a minimum of 5 users.
- This plan provides unlimited files, contributors, viewers, and call recordings.
- It is a solid choice for companies with multiple teams that need to collaborate effectively on user research projects.
Enterprise Plan
- This plan offers custom pricing and is designed for large organizations.
- It includes all the features from the Standard Plan, plus additional security, compliance (such as HIPAA, GDPR, and SOC2), and extra support.
- There are no platform limitations in this plan.
Key Features Across Plans
- AI-Powered Analysis: Features like automated tagging, notes, and AI-driven data analysis to identify patterns are available across most plans.
- Transcription: Automatic transcription in over 40 languages is included.
- Privacy and Compliance: Tools for protecting personally identifiable information (PII) such as face blurring, voice garbling, and data compliance are part of the higher-tier plans.
- Live Notes and Streaming: Live notes and the ability to live stream interviews to stakeholders are available in the paid plans.
Additional Notes
- Marvin does not offer a month-to-month payment option; plans are typically billed annually.
- The customer support team is known for being responsive and helpful.
Marvin - Integration and Compatibility
Marvin: An AI Toolkit for Natural Language Interfaces
Marvin, as an AI toolkit for building natural language interfaces, integrates and operates across various platforms and tools in several key ways:
Integration with AI Models and Services
Marvin 1.1 supports multiple AI model providers, including OpenAI, Anthropic, and the Azure OpenAI Service. This is achieved through a new providers interface, allowing developers to switch between these services by setting the `llm_model` or the `MARVIN_LLM_MODEL` environment variable. This flexibility ensures that Marvin can be used with different large language models (LLMs) as long as they are compatible with the OpenAI functions API.
Compatibility with Other Tools and Frameworks
Marvin is built to be highly compatible with existing development tools and frameworks. It uses familiar interfaces like Pydantic models, enums, and functions, making it easier for developers to integrate Marvin into their existing software projects. For example, Marvin’s AI Models, AI Classifiers, and AI Functions can be used in conjunction with other libraries and frameworks, allowing developers to bring AI capabilities into traditional software projects with minimal additional code.
Multi-Modal Support
Marvin supports multi-modal interactions, including image and audio generation, as well as using images as inputs for extraction and classification. This multi-modal capability makes Marvin versatile and adaptable to various application scenarios.
State Management and Application Integration
Marvin applications blend natural language interfaces with structured, privately managed states. These applications are built on top of Marvin’s assistants API, which ensures they inherit all the functionality of assistants. This approach streamlines user interaction and ensures compatibility with more structured, conventional applications. Marvin’s ability to call tools enhances its functionality, bridging the gap between natural language inputs and the specific requirements of traditional applications.
Parallel Processing and Streaming
Marvin 1.1 introduces a `.map()` method for AI Models, AI Classifiers, and AI Functions, allowing them to process multiple inputs concurrently. This parallel processing capability, along with the support for streaming outputs, makes Marvin more efficient and suitable for real-time applications.
Conclusion
While the specific resources provided do not detail extensive integrations with a wide array of third-party tools beyond AI model services and general development frameworks, Marvin’s design ensures it can be easily integrated into various software projects, making it a versatile tool for developers.
Marvin - Customer Support and Resources
Customer Support
Marvin places a strong emphasis on customer support. The platform is known for its responsive and attentive customer service. If users encounter any issues, they can rely on Marvin’s customer support team to resolve their problems promptly. This is highlighted as a key strength, as the team is committed to ensuring customer satisfaction.
Additional Resources
Documentation and Guides
Marvin provides comprehensive documentation, including a Getting Started guide, a Cookbook, and detailed Docs. These resources help users get familiar with the platform and its various functions.
API Reference
For developers, Marvin offers an API Reference that outlines how to integrate and use Marvin’s AI functions within their own applications.
Community Support
Users can benefit from a community support system, where they can interact with other users, share experiences, and get help from the community.
Tutorials and Examples
The platform includes tutorials and examples that demonstrate how to use Marvin’s functions effectively. For instance, the tutorial section shows how to generate synthetic data, perform sentiment analysis, and more.
Custom AI Functions
Marvin allows users to create custom AI functions that can be integrated into their Python code. These functions leverage Large Language Models (LLMs) to generate outputs based on the function’s description and inputs, making it easier to handle complex tasks without writing extensive source code.
Integrations and Compatibility
Marvin integrates well with other tools and platforms, such as Zoho One, which can be particularly useful for enterprises. This integration ensures that users can complement their existing tool stack without disrupting their workflow.
By providing these support options and resources, Marvin ensures that users have the necessary tools and assistance to maximize the benefits of its AI-driven data tools.
Marvin - Pros and Cons
Advantages of Marvin
Marvin, particularly in its application as a qualitative data analysis platform and research repository, offers several significant advantages:Centralized Data Management
Marvin provides a centralized location to store and organize all user research data, including interviews, transcripts, notes, quotes, and reports. This makes it easier to access and manage large volumes of data.AI-Powered Analysis
The platform uses AI for transcription, tagging, and pattern recognition, which significantly enhances the analysis process. It can produce accurate and editable transcripts in minutes, even in over 40 languages and dialects.Collaborative Features
Marvin facilitates collaborative research with features like shared access to the research repository, the ability to create playlists of key customer quotes, and livestreaming interviews to stakeholders. This improves teamwork and stakeholder engagement.Data Security and Privacy
Marvin includes robust data security features, such as automatic removal of personal information from interview transcripts, face blurring, and voice garbling. This ensures sensitive data is protected.Time-Saving Features
The platform offers time-saving features like automatic note-taking, AI-powered search, and comprehensive AI analysis of survey results. These features help researchers focus on high-value tasks like spotting patterns and generating insights.Interactive Reporting
Marvin allows users to create shareable clips, highlight reels, and interactive reports on insights uncovered, making it easier to present findings to stakeholders.Disadvantages of Marvin
Despite its numerous benefits, Marvin also has some notable drawbacks:Steep Learning Curve
Marvin has a specific way of doing things, which can lead to a steep learning curve. Users may find the tagging taxonomy confusing, especially for thematic analysis, and the interface can be cluttered with information scattered across different screens.Pricing
The pricing tiers of Marvin are not friendly to small businesses or those in seasonal industries, as it does not offer a month-to-month payment option and requires a minimum of 5 users for paid plans.AI Summarization Issues
Some users find the AI summary notes to be hit-or-miss, and the AI’s tendency to summarize everything automatically can be annoying. Users often prefer to use the ‘Ask AI’ feature manually.Manual Processes
There are some manual processes, such as uploading surveys, which can be time-consuming and less efficient compared to automated processes.Limited Customization
In its application as a lightweight AI toolkit, Marvin has limited customization options for AI models, which can be a challenge for users needing more flexibility in their AI configurations. By considering these points, users can make a more informed decision about whether Marvin aligns with their needs and workflow. Marvin - Comparison with Competitors
When Comparing Marvin with Competitors
When comparing Marvin, an AI-powered research assistant, with its competitors in the data tools and AI-driven product category, several key features and differences stand out.
Unique Features of Marvin
- Automated Tagging and Notes: Marvin automatically tags and takes notes during interviews and meetings, helping researchers identify patterns they might have missed.
- Centralized Research Repository: It provides a centralized platform to store, search, and share all user research data, including interviews, transcripts, notes, and surveys.
- AI-Driven Data Analysis: Marvin uses AI to analyze data and identify patterns, and it supports over 40 languages for transcription.
- Compliance and Security: Marvin complies with HIPAA, GDPR, and SOC2, ensuring data protection and privacy filters.
- Shareable Reports and Video Clips: Users can create and share reports and video clips, making it easier to communicate insights to stakeholders.
Alternatives and Their Key Features
Discuss.io
- Live Video Interviews and Mobile Screen Sharing: Discuss.io allows for live video interviews and mobile screen sharing, along with interactive whiteboards and powerful tagging features. It is particularly useful for global brands and agencies conducting qualitative research.
- Highlight Reels and Market Research Services: Discuss.io helps in creating highlight reels and offers additional services like moderation, human translations, recruiting, and program management.
Dovetail
- Qualitative Data Analysis: Dovetail is focused on analyzing data from interviews, usability testing, and survey responses. It allows users to identify patterns using a variety of qualitative research methods and transform qualitative data into quantitative data.
- Intuitive Controls and Global Tags: Dovetail offers drag-and-drop controls and the ability to add global tags, making it easy to organize and visualize data.
Condens
- AI-Driven Analysis and Multilingual Transcription: Condens streamlines user research with AI-driven analysis and multilingual transcription. It enhances collaboration and accessibility across teams and provides enterprise-ready solutions with scalability and security.
- Collaborative Tools: Condens offers collaborative tools to transform raw data into actionable insights and integrates seamlessly with existing workflows.
InsightLab
- AI-Powered Interviewing and Thematic Analysis: InsightLab uses AI for dynamic, conversational surveys that adapt to participant responses. It provides automated transcripts, thematic analyses, and handles data in over 70 languages.
- Scalable Qualitative Analysis: InsightLab is designed for efficient and scalable qualitative analysis, helping product teams make informed decisions by uncovering customer behavior.
UserBit
- Unified Workspace and Analytical Tools: UserBit offers a unified workspace for organizing user interviews, feedback, audio, videos, and notes. It includes powerful analytical tools like tagging and affinity diagrams to transform raw data into actionable insights.
- UX Tools: UserBit provides specialized UX tools for creating and exporting design outputs such as journey maps, visual sitemaps, and personas.
Other Notable Alternatives
Usersnap
- User Feedback Platform: Usersnap is a platform for product teams to gather product-specific insights, accelerate user testing, and improve stakeholder feedback loops. It is known for capturing issues with visuals and automating tech data to improve QA speed and cross-team communication.
Insight7
- AI-Driven Insights: Insight7 uses AI to extract themes, sentiments, and insights from data quickly. It supports various file formats and helps in prioritizing problems based on key signals, saving time and money on research and development.
Convo
- AI-Moderated Interviews: Convo is an AI-moderated qualitative user research platform that allows for real-time interviews at the scale and scope of surveys. It automatically re-analyzes data with new responses, ensuring up-to-date insights.
Each of these alternatives offers unique features that cater to different needs in the realm of qualitative data analysis and user research. While Marvin excels in automated tagging, centralized repositories, and compliance, other tools like Discuss.io, Dovetail, Condens, InsightLab, UserBit, Usersnap, Insight7, and Convo provide specialized functionalities that might be more suitable depending on the specific requirements of the user.
Marvin - Frequently Asked Questions
What is Marvin and how does it work?
Marvin is an AI-powered tool that utilizes OpenAI models to provide various functionalities such as classification, generation, and custom AI functions. To use Marvin, you need an OpenAI API key, which can be created on the OpenAI platform. Marvin integrates these models into Python functions, making it easy to perform tasks like sentiment analysis, intent classification, and data generation.
What are the main features of Marvin?
Marvin offers several key features:
- Classification: Allows you to classify text into predefined labels, useful for tasks like sentiment analysis and intent classification.
- Generation: Enables the creation of synthetic data from schemas or natural language instructions, which is useful for ideation, testing, and data augmentation.
- AI Functions: Lets you create custom AI-powered behaviors by combining inputs, instructions, and output types, similar to regular Python functions but generated by large language models (LLMs) on-demand.
How do I use Marvin for classification tasks?
To use Marvin for classification, you can call the classify function and provide the text you want to classify along with a list of labels. For example, you can classify text as having a positive or negative sentiment by passing the text and a list of sentiment labels to the classify function.
What is included in the free trial of Marvin?
Marvin offers a 14-day free trial that includes all the functionality of the selected plan. During the trial, you can use the service with a single user, and subsequent users will be charged immediately. This trial allows you to test whether Marvin is the right tool for your needs.
How does Marvin’s pricing work?
Marvin is offered on a software-as-a-service (SaaS) basis via a subscription model. You can pay either monthly or annually, with a 10% discount on annual plans. The pricing is per user, and there are several tiers (Starter, Professional, Business, and Enterprise) that cater to different needs and offer varying levels of features and support.
Are there any additional fees or hidden charges with Marvin?
No, there are no hidden charges or surprise fees with Marvin. The pricing plans are all-inclusive, covering the specified features and services. Any optional add-ons or services that may incur extra charges are clearly communicated before you decide to use them.
How can I use Marvin for generating synthetic data?
You can use Marvin’s generate function to produce synthetic data. This function takes either a target type or natural language instructions (or both) and the number of items to generate, returning a list of synthetic data that complies with the instructions. This is useful for tasks like data augmentation, testing, and populating databases.
Can Marvin be used for custom AI tasks?
Yes, Marvin allows you to create custom AI functions that can handle complex tasks. These functions look like regular Python functions but are generated by LLMs on-demand, enabling you to perform tasks that would otherwise require complex models and extensive training data.
Is there support available for Marvin users?
Yes, Marvin provides support for its users. The level of support varies by plan, with more comprehensive support available in the higher-tier plans. Additionally, local currency billing and mobile app access are available to make the service more accessible and user-friendly.
Can Marvin be integrated with other tools and platforms?
Marvin can be integrated with various tools and platforms. For example, there is a version of Marvin specifically designed to work with ZOHO CRM, helping users navigate and manage their CRM tasks more efficiently.
Marvin - Conclusion and Recommendation
Final Assessment of Marvin in the Data Tools AI-Driven Product Category
Marvin is a comprehensive qualitative data analysis platform that leverages AI to streamline and enhance research processes. Here’s a detailed assessment of who would benefit most from using Marvin and an overall recommendation.
Key Features and Benefits
- Centralized Data Management: Marvin provides a centralized location to store and organize all user research data, including interviews, transcripts, notes, quotes, and reports. This is particularly beneficial for researchers who need to consolidate large volumes of data in one place.
- AI-Powered Analysis: The platform offers advanced AI-powered tools for transcription, tagging, and pattern recognition. It can automatically take notes during interviews, find patterns that might be missed, and tag important labels for later investigation.
- Collaborative Research: Marvin facilitates collaborative research with features like shared access to the research repository, the ability to create playlists of key customer quotes, and live streaming of interviews to stakeholders.
- Data Protection: The platform includes secure data management features such as automatic removal of personally identifiable information (PII) from interview transcripts, face blurring, and voice garbling.
- Insight Generation and Sharing: Marvin allows users to query their entire research repository using an AI-powered search engine and generate shareable clips, highlight reels, and interactive reports on insights uncovered.
Who Would Benefit Most
Marvin is highly beneficial for researchers, particularly those in user experience (UX) research, market research, and any field requiring the analysis of large volumes of qualitative data. Here are some specific groups that would benefit:
- UX Researchers: Marvin’s ability to organize and analyze user interviews, transcripts, and notes makes it an invaluable tool for UX researchers looking to extract meaningful insights and patterns from user data.
- Market Researchers: The platform’s AI-powered analysis and data visualization capabilities are ideal for market researchers who need to generate insights from customer feedback and other qualitative data.
- Academic Researchers: Researchers in various academic fields can benefit from Marvin’s centralized data management, automatic note-taking, and AI-driven pattern recognition features.
Overall Recommendation
Marvin is a powerful tool for anyone involved in qualitative data analysis. Here are some key points to consider:
- Ease of Use: While Marvin offers a wide range of features, it has a steep learning curve. Users need to invest time in learning how to use the platform effectively.
- Cost: The pricing tiers may not be friendly for startups or small teams, as the paid plans require a minimum of 5 users.
- AI Accuracy: The AI summary notes can be hit-or-miss, and some users may find the tagging taxonomy confusing for thematic analysis.
Despite these challenges, Marvin’s benefits in terms of data organization, AI-powered analysis, and collaborative features make it a valuable asset for researchers. If you are willing to invest the time to learn the platform and can afford the pricing, Marvin can significantly enhance your qualitative data analysis capabilities.
In summary, Marvin is a strong choice for researchers looking to streamline their qualitative data analysis processes, especially those in UX and market research. However, it is important to be aware of the potential learning curve and cost implications.