Content Type: Starter App

Categories: Starter Apps,AWS,Artificial Intelligence

<h3><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.85);">Configuration</span></h3><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">All modules that are needed to interact with Mendix Cloud GenAI Resource Packs, an LLM from (Azure) OpenAI or Amazon Bedrock are already installed. The app contains functionalities to configure OpenAI or Amazon Bedrock out of the box. Feel free to add your own models or remove the existing ones.</span></p><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Read this </span><u style="color: rgb(6, 74, 228); background-color: transparent;"><a href="https://docs.mendix.com/appstore/modules/genai/how-to/starter-template/" rel="noopener noreferrer" target="_blank">how-to guide</a></u><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);"> for more information about setting up access to a model provider.</span></p><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Before users can chat with a model, the admin needs to create Bot Configuration(s) for users to select in the chat interface.</span></p><ol><li><ol style="list-style-type: lower-alpha"><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Display&nbsp;name:&nbsp;what&nbsp;users&nbsp;will&nbsp;see&nbsp;on&nbsp;the&nbsp;page</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Is&nbsp;Selectable&nbsp;in&nbsp;UI:&nbsp;decide&nbsp;if&nbsp;a&nbsp;configuration&nbsp;should&nbsp;be&nbsp;selectable</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Model&nbsp;selection:&nbsp;models&nbsp;of&nbsp;a&nbsp;specific&nbsp;provider&nbsp;that&nbsp;were&nbsp;configured&nbsp;before</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Action&nbsp;microflow:&nbsp;select&nbsp;what&nbsp;action&nbsp;microflow&nbsp;should&nbsp;be&nbsp;executed.&nbsp;The&nbsp;provided&nbsp;one&nbsp;(&quot;ChatContext_ChatWithHistory_ActionMicroflow&quot;)&nbsp;supports&nbsp;all&nbsp;providers.&nbsp;Feel&nbsp;free&nbsp;to&nbsp;customize&nbsp;the&nbsp;microflow&nbsp;to&nbsp;your&nbsp;needs</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">One&nbsp;or&nbsp;multiple&nbsp;Knowledge&nbsp;Base&nbsp;Retrieval&nbsp;tools&nbsp;can&nbsp;be&nbsp;added&nbsp;to&nbsp;a&nbsp;BotConfig&nbsp;from&nbsp;the&nbsp;overview&nbsp;page.&nbsp;This&nbsp;enable&nbsp;the&nbsp;chat&nbsp;to&nbsp;be&nbsp;enriched&nbsp;with&nbsp;specific&nbsp;knowledge</span></li></ol></li></ol><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);"> </span></p><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Additionally, you can create starting points for prompt engineering, that will be available to your users. Both can be configured at runtime (or in the After Startup):</span></p><ol><li><ol style="list-style-type: lower-alpha"><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Instructions:&nbsp;a&nbsp;prefilled&nbsp;instruction&nbsp;to&nbsp;be&nbsp;selected&nbsp;in&nbsp;the&nbsp;chat&nbsp;interface.&nbsp;Instructions&nbsp;augment&nbsp;the&nbsp;&quot;System&nbsp;Prompt&quot;&nbsp;and&nbsp;influence&nbsp;the&nbsp;model&#39;s&nbsp;behavior.&nbsp;Users&nbsp;can&nbsp;create&nbsp;their&nbsp;own&nbsp;custom&nbsp;instructions&nbsp;as&nbsp;well&nbsp;(only&nbsp;visible&nbsp;to&nbsp;them).</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Initial&nbsp;User&nbsp;Prompts:&nbsp;buttons&nbsp;that&nbsp;can&nbsp;be&nbsp;clicked&nbsp;in&nbsp;new&nbsp;chats&nbsp;by&nbsp;the&nbsp;user&nbsp;to&nbsp;prefill&nbsp;the&nbsp;user&nbsp;prompt&nbsp;and&nbsp;a&nbsp;corresponding&nbsp;instruction.</span></li></ol></li></ol><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);"> </span></p><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">To give you an idea, a few examples are automatically created for both. Those can be deleted at runtime and / or deactivated in the ASU microflow if not needed:</span></p><ol><li><ol style="list-style-type: lower-alpha"><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Summarize&nbsp;my&nbsp;meeting&nbsp;notes:&nbsp;inserts&nbsp;a&nbsp;sample&nbsp;text&nbsp;of&nbsp;meeting&nbsp;notes&nbsp;and&nbsp;instructs&nbsp;the&nbsp;LLM&nbsp;to&nbsp;find&nbsp;important&nbsp;information&nbsp;(deadlines,&nbsp;action&nbsp;points).&nbsp;The&nbsp;LLM&nbsp;should&nbsp;then&nbsp;summarize&nbsp;the&nbsp;text&nbsp;using&nbsp;headers&nbsp;and&nbsp;bullet&nbsp;points.</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Launch&nbsp;an&nbsp;AI&nbsp;Chatbot&nbsp;internally:&nbsp;instruct&nbsp;the&nbsp;LLM&nbsp;to&nbsp;create&nbsp;an&nbsp;engaging&nbsp;text&nbsp;for&nbsp;the&nbsp;target&nbsp;audience.&nbsp;The&nbsp;text&nbsp;should&nbsp;describe&nbsp;the&nbsp;technology&nbsp;and&nbsp;business&nbsp;impact.</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Launch&nbsp;an&nbsp;AI&nbsp;Chatbot&nbsp;on&nbsp;LinkedIn:&nbsp;the&nbsp;LLM&nbsp;should&nbsp;create&nbsp;an&nbsp;engaging&nbsp;LinkedIn&nbsp;post&nbsp;with&nbsp;hashtags&nbsp;and&nbsp;emojis&nbsp;about&nbsp;a&nbsp;recently&nbsp;launched&nbsp;AI&nbsp;chatbot.</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Help&nbsp;me&nbsp;decide&nbsp;launching&nbsp;an&nbsp;MVP&nbsp;product:&nbsp;the&nbsp;LLM&nbsp;helps&nbsp;you&nbsp;to&nbsp;brainstorm&nbsp;and&nbsp;prepare&nbsp;for&nbsp;discussions&nbsp;about&nbsp;your&nbsp;MVP&nbsp;product.</span></li></ol></li></ol><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);"> </span></p><h3><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.85);">Customize&nbsp;AI&nbsp;Bot</span></h3><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">This app serves as a starting point and there are many ways to customize your app:</span></p><ol><li><ol style="list-style-type: lower-alpha"><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Add&nbsp;your&nbsp;custom&nbsp;styling&nbsp;(see&nbsp;</span><u style="color: rgb(6, 74, 228); background-color: transparent;"><a href="https://docs.mendix.com/howto/front-end/customize-styling-new/" rel="noopener noreferrer" target="_blank">Customize&nbsp;Styling</a></u><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">&nbsp;for&nbsp;more&nbsp;details)</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Customize&nbsp;the&nbsp;initial&nbsp;user&nbsp;prompts&nbsp;that&nbsp;are&nbsp;suggested&nbsp;in&nbsp;new&nbsp;chats</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Redesign&nbsp;the&nbsp;chat&nbsp;page&nbsp;to&nbsp;your&nbsp;needs</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Add&nbsp;custom&nbsp;pre-&nbsp;or&nbsp;post-processing&nbsp;logic&nbsp;into&nbsp;the&nbsp;action&nbsp;microflow&nbsp;that&nbsp;interacts&nbsp;with&nbsp;the&nbsp;LLM,&nbsp;for&nbsp;example&nbsp;to&nbsp;bring&nbsp;your&nbsp;own&nbsp;knowledge&nbsp;base.</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Add&nbsp;your&nbsp;own&nbsp;LLM&nbsp;provider&nbsp;connector</span></li></ol></li></ol><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);"> </span></p><h3><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.85);">Document&nbsp;chat</span></h3><h4><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.85);">Mendix&nbsp;Cloud&nbsp;or&nbsp;Amazon&nbsp;Bedrock</span></h4><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">When using Mendix Cloud or Amazon Bedrock as a provider, users can upload up to five documents of various types to let the model analyze and base the response on those.</span></p><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);"> </span></p><h4><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.85);">(Azure)&nbsp;OpenAI,&nbsp;Mistral&nbsp;and&nbsp;Google&nbsp;Gemini</span></h4><p><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">When selecting (Azure) OpenAI, Mistral or Google Gemini, only a single PDF can be passed. The text is extracted from the PDF and inserted into the prompt. This does mean that:</span></p><ol><li><ol style="list-style-type: lower-alpha"><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Graphics&nbsp;(this&nbsp;might&nbsp;include&nbsp;scans)&nbsp;and&nbsp;formatting&nbsp;from&nbsp;the&nbsp;PDF&nbsp;will&nbsp;mostly&nbsp;be&nbsp;ignored.</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Long&nbsp;PDFs&nbsp;can&nbsp;reduce&nbsp;the&nbsp;accuracy&nbsp;of&nbsp;responses&nbsp;(due&nbsp;to&nbsp;distraction&nbsp;from&nbsp;the&nbsp;important&nbsp;content)&nbsp;or&nbsp;even&nbsp;cause&nbsp;an&nbsp;error&nbsp;when&nbsp;making&nbsp;a&nbsp;request&nbsp;(due&nbsp;to&nbsp;exceeding&nbsp;the&nbsp;model&#39;s&nbsp;context&nbsp;window).&nbsp;A&nbsp;warning&nbsp;is&nbsp;shown&nbsp;to&nbsp;the&nbsp;user&nbsp;when&nbsp;a&nbsp;long&nbsp;PDF&nbsp;is&nbsp;being&nbsp;used.</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">As&nbsp;the&nbsp;PDF&#39;s&nbsp;content&nbsp;is&nbsp;passed&nbsp;along&nbsp;the&nbsp;request,&nbsp;the&nbsp;token&nbsp;usage&nbsp;is&nbsp;increased&nbsp;(and&nbsp;thus&nbsp;costs)</span></li><li><span style="background-color: rgb(255, 255, 255); color: rgba(10, 19, 36, 0.7);">Not&nbsp;every&nbsp;model&nbsp;performs&nbsp;equally&nbsp;well&nbsp;with&nbsp;PDF&nbsp;content&nbsp;(for&nbsp;example,&nbsp;OpenAI&#39;s&nbsp;GPT-4o&nbsp;performs&nbsp;better&nbsp;than&nbsp;GPT-3.5&nbsp;turbo)</span></li></ol></li></ol><p>&nbsp;</p>