Publishing GenAI apps in 10 minutes using GPT Builder and AWS PartyRock

I just published the GenAI Advisor app on GPT Store. It only took 10 minutes of tinkering! I also took similar time to launch a GenAI app using AWS PartyRock. Awesome results compared below.

GenAI Advisor on GPT Builder

I had five turns of conversation with GPT Builder on what I intend to create. Selected a logo. Did some tweaking of the instruction or prompt generating the GPT. Refined the recommended sample prompts helping first time users. Tested the GPT a few times. Published to the world! Here is what the finished product looks like.

You can try out the new GenAI Advisor here (requires ChatGPT plus). Try the sample prompts or ask away your custom prompts. I have an Easter Egg embedded in every response, hope you like it! If you do not have a Plus account, no worries, I will share my journey next.

Building the GPT

I derived the instruction prompt from the mock design and information flow I did in prior articles. I chose the default GPT settings and only tweaked the recommended conversation starters.

Next step is trying out the GPT. So, as a user of the GPT I tried one of the sample prompts. How is Meta's recent research advancing GenAI?

The response used search tool to find relevant retrieval sources including the Jan 2024 article posted on the Engineering at Meta website titled How Meta is advancing GenAI. This is a nice coincidence as I did not know when refining the sample prompt, that GPT Builder recommended, that it matches the article title so closely.

First part of the response pretty much resembles asking ChatGPT the same prompt, with the difference in logo suggesting GenAI Advisor is responding.

However, here is what is unique about the response. Based on the response generated, I have instructed GenAI Advisor to generate a thematic visual at the end of the response.

Now I can generate part of the research for future articles of AI for Everyone using GenAI Advisor including the featured image, using one prompt. I already feel more productive and smarter!

Testing the GPT

Ok, all this seems too good to be true, right? So here comes the reality check.

The next prompt I tried was a repeat of what I used for the GenAI Advisor prototype I built using Amazon Q in my prior article.

Prompt: List 20 quantitative insights relating to generative AI impacting future of work

The response including a disclaimer: Here's a list of 20 potential insights. These insights are hypothetical and based on trends observed in the industry, demonstrating the wide-ranging impact of GenAI on the future of work.

Improving the GPT

This is not what I intended for GenAI Advisor to deliver. I had to tweek the GPT instruction with the following.

GPT Instruction: Do not generate hypothetical responses. Only respond when certain about accuracy of insights from reputable sources you can cite.

This yielded a much better response to the same prompt. I also noted that the sources cited followed my recommended sources in the instruction to the GPT, like Statista and McKinsey.

Publishing on GPT Store

Publishing GenAI Advisor is a single click action. Here is how the GenAI Advisor looks when discovering it in GPT Store. Typing "latest trends" seems to list GenAI Advisor as top recommendation. Not bad for a 10 minute speed launch!

What's next for GenAI Advisor on GPT Builder

While it is exciting to play with this new app store and GPT Builder tool, I am left with following questions and actions which I will leave in a parking lot for future articles.

1. How do I add custom format to responses, like tables and charts?
2. What are the limits of the instructions prompt?
3. What are the best practices for the instructions prompt?
4. When should I add knowledge retrieval documents?
5. How do I make sure the retrieval documents are used in response, prioritized over search tool or over ChatGPT's own response?
6. How do I add custom actions?
7. How do I optimize discovery of the GPT on GPT Store search?

GenAI Advisor on AWS PartyRock

Time to switch gears and compare what similar insutruction prompt and effort yields on AWS PartyRock. Here is what GenAI Advisor built on AWS PartyRock looks like. You do not need any subscription to either get started building the app or accessing other apps. Only a simple signup would get you there.

GenAI Advisor on PartyRock not only looks very different, the builder experience was very unique and interesting as well.

Benefits from Amazon Bedrock

AWS PartyRock sits on top of Amazon Bedrock. Here is a brief description of Amazon Bedrock from the official website.

Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs) from leading AI companies like AI21 Labs, Anthropic, Cohere, Meta, Stability AI, and Amazon via a single API, along with a broad set of capabilities you need to build generative AI applications with security, privacy, and responsible AI.

This means AWS PartyRock enables a choice of LLMs to build GenAI apps. I could not only select a text-to-text model suitable for my app but also an image generation model for creating thematic images. I can also tune model specific parameters like Temperature and Top P to match my app requirements.

Configurable widgets and layout

Another cool capability of AWS PartyRock is the configurable and drag-drop frindly widgets. I could configure a user input widget, connect it to the LLM response widget, and connect the LLM response to the image generation widget, without writing a single line of code. This makes the GenAI Advisor look more like a custom app than like just another chatbot.

The image generation widget is also configurable for style presets including 3D model, line art, anime, photography, etc. I chose photography style preset to generate realistic images to match the response.

What's next for GenAI Advisor on AWS PartyRock

Here are my set of actions for further explorign AWS PartyRock.

1. How to add more widgets and types to build a more complex user experience?
2. What interesting patterns (like chain of thought) can be explored when connecting multiple widget inputs/outputs?
3. What benefits can I deliver by combining multiple models in same app?
4. How to navigate the limitation of search tool integration?
5. How to improve discovery of GenAI Advisor on AWS PartyRock / Discover?

Now I have three GenAI Advisor apps built using Amazon Q, AWS PartyRock, and GPT Builder. Each has a different technology stack, user interface, builder experience, models, tools integration, and data security considerations. Depending on your specific use case you may decide to follow one of these paths for your GenAI journey. If you are considering building with sensitive enterprise data then Amazon Q may offer the most granular security and access controls. If you want to benefit from the latest OpenAI models then GPT Builder might be your path. If you want to rapidly explore a unique and flexible user interface using a no code builder experience then AWS PartyRock might be your preference.

Next in series, I am planning to explore open source models available on Hugging Face. Plus I want to continue learning more about Amazon Q, GPT Builder, and AWS PartyRock. What would you like to read more about? Please comment or DM.

The author writes about generative AI to share his personal interest in this rapidly evolving field. The author's opinions are his own and do not represent the views of any employer or other entity with which the author may be associated.

Thanks for reading the AI for Everyone newsletter. Share and subscribe for free to receive new posts and support my work.