GAIP Release notes

February 18, 2025

New feature

• Project-wise member permissions: Project-specific user access permissions are now supported. Account Owners can manage access by adding or removing members for specific projects from the Settings page. By default, members do not have access to any projects. They can only see and access projects they have created or have been granted access to.

Action Required

• With this release, project-specific user access permissions are now enforced. Account Owners must manually grant access to members for each project. By default, members will not have access to any projects unless explicitly added via the Settings page.

February 17, 2025

New Feature

• Embedding model selection from UI: Maira Embedding models can now be selected from the UI for each project during setup and can be updated later.
• LLM model provider selection from UI: The LLM Model Provider, such as OpenAI, Anthropic, or Azure, can now be selected. Users can choose a model from the selected provider in their profile. This allows to create profiles with different providers to manage cases where one provider may be unavailable or cater to specific needs.
• Real-Time Call Capture & Processing in Maira: Maira now supports registering phone/VoIP calls, capturing them in real-time, and converting them to text. Using our profile feature, these transcriptions can be automatically processed into summaries, reports, and other structured outputs, enhancing efficiency and insights.

February 16, 2025

New Feature

• Azure Integration with Maira: Maira now supports OpenAI-provided LLMs deployed in Azure, expanding deployment flexibility.

• New Keyword Extraction Model:

  • A new Vietnamese and English keyword extraction model has been implemented to address performance delays caused by the previous model.

Enhancements

• items/search Endpoint Speed Improvement:
  • Added is_context_aware in search settings. When set to false, significantly improves search speed.
  • Introduced an optional language parameter in the request body. We recommend passing this parameter to further enhance response speed.

Deprecations & Migrations

• Endpoint Deprecation and Migration:
  • gpt/ask and gpt/ask/vision are now deprecated and replaced with maira/ask and maira/vision.
  • The deprecated endpoints will remain supported temporarily until client-side migration is complete, after which they will be removed.

February 4, 2025

New Feature

• Filter Items with price 0 in v1/items/search: Now you can filter out items with a price of 0 in the items/search endpoint. Previously the minimum price was 1. Now it is possible to search for items with price 0 (free items).

February 2, 2025

• Acknowledge_only in GPT Ask: Introduced the acknowledge_only flag in the v1/gpt/ask endpoint. When building AI agents, not all inputs require an immediate response—sometimes, we simply want the agent to listen and retain context for future interactions. This flag enables that functionality, improving contextual continuity in conversations.

January 28, 2025

Enhancements

• Easier Maira Profile creation
  • The Randomness option is now available in the profile creation UI. This allows users to generate sample text in the introduction, specifying whether the LLM should use only the provided context or include general knowledge as well.
  • Previously, this selection was part of the project creation flow, affecting all profiles in the project. This has now been removed.
  • It was also available in project settings, where it affected all profiles in the project. This has now been removed.
• User Re-invitation
  • Account admins can now re-send invitations to previously invited members.
  • This helps address scenarios like email delivery failures or other unexpected issues.
• New UI features in Maira playground
  • Conversation History Access: Users can now access their past conversation history directly from the UI.
  • Enhanced Conversation Details: The conversation details page now matches the layout of GAIP conversation details, allowing users to view evaluation scores and top references easily.
  • Response Insights: Response time and the profile used to generate each response are now displayed directly under the response for better transparency.
  • Simplified Profile Switching: Switching profiles has been made more intuitive and user-friendly.
  • Voice Input Support: Users can now ask questions using their device’s microphone for a seamless experience.

Bug fix

• Repeat users table in insights page for personalizer projects, is no longer broken.

January 27, 2025

Enhancements

• Improved the Japanese language translation for better UX for the Japanese users.

January 09, 2025

Enhancements

• Previously, Conversation details page used to open as a modal. Now Conversation details will open as a new page when user click the "eye" button on a conversation history table. This link can be shared with others to grant access to the page.

December 29, 2024

Enhancements

• Improved item training pre-processing to handle non-English and non-Japanese languages, improving CSV alignment issues caused by special characters.

December 26, 2024

Enhancements

• Insights page tables now show member_ids along with user_id for improved Insights.

December 24, 2024

Enhancements

• Questionnaire logs detail page, now shows list of excluded items.

December 12, 2024

Enhancements

• Safe Embedding Generation for Item Search: Ensured item search endpoint stability by implementing fallback mechanisms. If external services like OpenAI or Azure become unavailable, our system will continue to function without disruption.
• Enhanced Keyword and Number Handling in Maira Document Search: Separated the processing of keywords and numbers in Maira's document search, improving the accuracy and relevance of results for the GPT-powered ask endpoint.
• Contextual Understanding Improvement for Maira Ask: Enhanced Maira’s ability to manage context switching during conversations, resulting in more accurate and context-aware responses. This will significantly improve the quality of chat-based interactions.

December 10, 2024

New Feature

• A new evaluation feature is introduced to assess how accurate responses are against the dataset. Evaluation scores can be viewed on the conversation history detail page.

December 08, 2024

Enhancements

• Search settings in Catalog Mapper now supports list of prioritize_category
• The sort_by_priorities feature for "items search" is enhanced to better support non-English languages.
• A default PDF extractor is added. If a new PDF extractor is added, it will override the default.

December 06, 2024

New Feature

• PDF Upload : PDF files can now be uploaded as Maira datasets. (Refer to the tutorial for more details about this update.)

December 03, 2024

New Feature

• Download Maira Documents : All documents in a Maira dataset can now be downloaded as CSV files from the document details page.

December 01, 2024

Enhancements

• Questionnaire recommended items now include a score.
• On the Dataset page, text search for catalog items has improved accuracy.

Bug Fix

• A bug in Questionnaire feature's keyword enabled function, is now fixed

November 19, 2024

New Feature

• Filter Datasets by Tag: On the Maira Datasets page in GAIP, datasets can now be filtered by tags. Additionally, when retrieving datasets using GET /v1/gpt/datasets/{dataset_id}, filtering by dataset tags is now supported.

Bug Fix

• Fix an issue related to "Cancel" button on Update Settings view

November 07, 2024

New Feature

• Conversation History Download: Conversation history can now be easily downloaded as a file from the project insights page. The generated file is saved for 30 days for re-download without re-generating.
• Easier document List access: The document list is now accessible from the dataset preview modal.
• Edit Documents in Conversation Context: Documents can now be edited directly from the context list in conversation history.

November 01, 2024

Enhancement

• On GPT Profile, now floating point number from 0 ~ 1 with a step of 0.1 can be set for for stop_p, temperature, and frequency.

Bug Fix

• A bug related Maira datasets auto-training failure is now fixed.

October 28, 2024

Enhancements

• Conversation search query now use exact matching, instead of partial matching of the query.

October 21, 2024

New Feature

• Train Maira Dataset from dataset page: You can now train a Maira dataset directly from the dataset list page by clicking on the train button in the "Actions" column.
• Pagination Support in /v1/items/scroll API: Introduced a scroll_id parameter in the request body and a new next_scroll_id in the response body to facilitate pagination using the scroll_id.
• Latest GPT Models and Parameters added to UI: Updated the latest supported GPT models and parameters on the GPT profile page.

Enhancements

• Log Message improvement: Trimmed log messages in the Dataset Update Log section when they exceed 85 characters for better clarity.

Bug Fix

• search_max_token Default Value Fix: Fixed a bug on the profile update form where search_max_token defaults to 2500 even if set to 0.

October 5, 2024

New Feature

• Anthropic Support: In addition to OpenAI, Maira now supports Anthropic models such as claude-3-5-sonnet and claude-3-opus. These models can be used in the same way as OpenAI models.
• Brainwave Support: Maira now integrates LLAMA-based Brainwave implementation. Brainwave is Gigalogy's in-house LLM inference product.
• Supported Models List: The GET gpt/models endpoint now allows retrieval of all supported models.
• Simplified Model Management: New endpoints have been added to add, update, and delete models. This significantly reduces the time needed to integrate new models into the system.

Enhancement

• Exact Phrase Search in Datasets and Documents: When using the GET /gpt/datasets and GET /datasets/{dataset_id}/documents endpoints with the keyword parameter, the search now matches the full phrase exactly. Previously, partial matches would return loosely related documents.

October 3, 2024

New Feature

• Payload-Only Update for Item Training: The Personalizer now allows all fields of the item catalog, including date fields, to be filterable. Additionally, previously trained items can gain this filterability by setting is_only_payload_update=true in the /items/train endpoint.
• Date Range Filters: Date range filtering is now supported on four search and recommendation-related endpoints: /item/search, /user/recommend, /search/recommend, and /items/recommend. API users can apply filters based on any date field in the catalog to refine search or recommendation results.
• Full Phrase Search in Conversations Endpoint
  The POST /conversations endpoint now supports exact phrase matching via the search_text field. If the entire string provided matches any field in a conversation, that conversation will be included in the results.

Bug Fix

• Fix for Insights Last Updated Timestamp: For newly created projects, if there is no last update available, the timestamp now displays as "N/A" instead of the default 1970 date.
• Bug Fix for GPT Profile Tags Input: Previously, if a user added a keyword or tag and clicked the submit button without pressing enter, the input was not processed. This issue has been resolved, and inputs are now handled correctly. Additionally, we've added a message below each input field to guide users on how to properly add tags.

September 15, 2024

Bug Fix

• GET/questionnaire/logs endpoint was not handling request_id filtering correctly in some instances. This is fixed now.

September 9, 2024

New Feature

• Now catalog items can be seen in the UI under personalizer project's dataset menu.
• New Endpoint: v1/categories/recommend endpoint now now be used to recommend categories based on user behavior, enhancing personalized experiences.
• New Endpoint: v1/interactions endpoint now collects user interactions with various elements like buttons, URLs, forms, inputs, etc. It enables mapping the user journey in greater detail for deeper insights.

Enhancement

• For Datasets table in Personalizer project and and Documents lists table in Maira projects, can now be sorted by the created_at column. Keyword search and filtering via drop-down options are also supported.
• Profile ID has been added on the Profile details page. User are now able to copy it by clicking the copy button.

Bug Fix

• After updating a a project profile, when creating a new profile, the form used to populate with the last updated profile information. This is is fixed now.

August 21, 2024

Enhancement

• Support for GPT-4o Mini: Added support for the GPT-4o Mini model, expanding our capabilities.
• JSON Format Support in Profiles: Profiles now support responses in JSON format, providing more flexibility in data handling.

August 20, 2024

New Feature

• Dynamic Keyword N-Gram: The range of words (1 ~ 3) for keyword definitions can now be customized per project. Based on these settings, the model will identify the top keywords from a query, allowing for more flexible and accurate extractions.
• Adjustable Search Depth: Search depth is now dynamic, offering improved control over performance. This can now be adjusted in search settings to prioritize speed where necessary.
• Prioritized Key Feature: A new prioritize_key option in the search settings allows you to designate a key that will take precedence during search, improving the relevance of the results.

Enhancement

• Introduced New Keyword Extraction Model, enhancing accuracy across all supported languages, including English, Japanese, and Vietnamese.

August 19, 2024

New Feature

• Added bulk documents upload option in the project Maira Datasets's Documents view.
• Added CSV download options for top browsed, purchased, rated items, and top users in the project Insights view.
• Added new parameters named search_settings in the project Setup view.

Enhancement

• Start button for "Images train" is disabled now by default and is only enabled after the Items train successfully run once in the project Settings view.

Bug Fix

• Fixed the Train Type selection bug in the project Settings view, in Train History section.
• Fixed the Average Cart Size calculation bug in the project Insights view.
• Fixed Conversation History download file bug in the project Insights view.

August 12, 2024

Enhancements

• The POST v1/items/recommend endpoint will no longer throw an error if an item does not exist; it will return null instead.
• Maira Profiles now support max_search_token set to 0, enabling answer generation from the LLM's general knowledge without adding external context.

August 2, 2024

New Feature

• Each profile in Maira, now has an option to set an evaluation. When enabled, this will calculate the accuracy of the answers generated by Maira models. This feature helps user to trust the responses provided by Maira.

July 16, 2024

New Feature

• Maira datasets, Profile, and settings are now available in the GAIP UI for each personalizer project.
• "Last updated at" info added to the insight page.
• Improved localization for Japanese language.
• Client key can be used for all end-points now.
• Questionnaire log can now be seen in the project insight page.

July 14, 2024

New feature

• Now tags can be added to Maira datasets. This tags can be used to control which datasets should be used for which profile. This feature gives the flexibility to use one database to support multiple purposes.

June 18, 2024

New feature

• Added "dataset upload history" table to the "dataset" tab for Personalizer projects.
• In insights page, "member ID" of top users is added.

June 9, 2024

Enhancement

• User can now edit search settings in mapper from the GAIP UI.
• Top browsed items and top rated items in GAIP insight page now can be clicked to see detail.

May 24, 2024

New feature

• Maira projects can now be setup from GAIP platform UI.

May 23, 2024

Enhancements

• questionnaire/recommend now have flags option in result format. Now it is possible to filter recommendations by flags. Ex: filter out items that are not in stock.
• profile in now used in questionnaire/recommend to build the context better. This will improve the prompt and consequently better finding products for recommendation.

May 19, 2024

Enhancements

• Improved gpt/ask context feature.
• Search result candidate cluster size is dynamic now for "questionnaire recommend". It can be set in questionnaire mapper now to use it.
• "Search settings" added to "item mapper". Parameters such as is_keyword_enabled, top_k, search_size is removed from the request body.
• Personalized search now have a priority sort feature. Now it is possible to prioritize items from certain category if they exists in search result.

May 15, 2024

Enhancement

• OpenAI's latest model gpt-4o is now supported in GAIP. This will improve the speed and sentence structure of Japanese language for /gpt/ask end point response significantly.

Apr 19, 2024

New feature

• Questionnaire mapper's result_format now have a excludes key, that can be used to exclude specific keys from the prompt_sections.

Apr 13, 2024

New feature

• Conversation Summary Feature. Now it is possible to get conversation summary in daily unit. Units can be set according to the need.
• Conversation Summary is saved as a regular document and trains immediately for future interactions with Maira. This helps the model to use summaries as Long Time Memory while generating answers.

Apr 11, 2024

Enhancement

• Item recommendations now no longer returns null ids for projects that contains both personalizer and Maira data
• Now you can use the endpoint /v1/categories/keywordsto build category navigation or see which category/sub category has the maximum number of products in the catalog.

Apr 5, 2024

New Feature

• Questionnaire mapper now supports result format. Users can set up how they want to see the questionnaire result. Users can set up required categories and optional categories. Recommended items from POST /v1/questionnaire/recommend will be categorized based on that. Optional categories come with their inclusion condition as wel

Mar 27, 2024

New feature

• From the GPT-Flow History table, you can now select which data (Query, response, feedback) you want to include in the training for future conversations and in which dataset.

Mar 27, 2024

New feature

• Now POST /v1/gpt/ask API have multiple options to generate answer. This can be set as the value of the key conversation_type, question type does not consider any previous conversation history. chat type considers n previous conversation history while generating an answer. The length of the conversation is defined by the parameter chat_history_length of GPT settings.

Mar 26, 2024

New feature

• In the same project, /gpt/ask API can now answer from both personalizer catalog and Maira datasets. Previously it could handle only one of the data sources.
• Introducing "Profiles" for our Maira! You can now create and save multiple profiles, each with its own settings . Users no longer have to manually adjust Maira settings every time they switch between different tasks or needs.

Mar 20, 2024

New feature

• keyword extractor changed for japanese language. This is done to improve the quality of japanese language search capabilities. top_k can control how wide or controlled search a customer wants to do in terms of keywords.
• questionnaire/recommend end point now has a new optional parameter called recommend_count. If it is set, the value of this parameter takes priority over recommend_count variable in questionnaire mapper.

Mar 16, 2024

New feature

• gpt/ask and questionnaire/recommend endpoints, when providing response with product IDs, now can include any key that was set in the mapper.
• Now it is easier to set how many product questionnaire/recommend will give, with the new key "recommend_count" in the mapper.
• introducing top_k to items/search, gpt/ask, giving ability to controls the number of keywords to consider for document searching.

Bug Fix

• questionnaire/recommend endpoint can now take Null values.

Mar 3, 2024

New feature

• UI-Based Project Setup: Now set up Personalization Projects directly via the UI, including file uploads, data mapping, and model training.
• Training History: View detailed training history in Project Settings for improved tracking and insights.
• Manual Training: Initiate model training manually from the UI for greater control over your personalization strategies.

Jan 25, 2024

New feature

• Add an export button at the bottom of the GPT flow history table. Exports the Table as a CSV file for the specified time range
• GAIP sidebar improved: Solutions are accessible directly now.

Bug Fix

• Bug at trial project status and upgrade flow