Building a new application can be daunting, especially when it comes to designing the architecture. This post explores how to leverage the power of an AI diagram generator, like ChatUML, to brainstorm, visualize, and refine your system design. We'll use the example of building a "Chat to Document" AI program to illustrate the process.

Let's say we want to build an AI-powered application that allows users to convert chat conversations into structured documents. Where do we start? ChatUML can help!

Step 1: Clarification is Key​

Before diving into diagrams, it's crucial to provide the AI with a clear understanding of your project. Ask the AI assistant to follow up with as much clarifying questions as possible to ensure the it grasps the core functionality and requirements. This initial prompt sets the stage for a successful design process:

Help me create a tech design document for a project i'm working on.
Do not generate anything until you are crystal clear about what you need to know.
Feel free to ask me as much clarify questions as possible.

I want to create a tech desgin doc about a chat to document AI program.

After this, the AI assistant will start asking questions so you can provide information about your project. For example, in this case, I told the AI the following:

• This project is a system that allows users to upload their files. The files will get processed in the backend and stored in a vector DB.
• The technology I'm going to use is: Rust in the backend, React for the frontend, and Postgres with vector support. I didn't need to explicitly tell the AI how all the components interact with each other, as I want it to figure that out for me.
• There is a requirement that the user should not be able to access files that are not theirs.

Step 2: System Overview​

After a few rounds of discussion, the AI gained a pretty clear understanding of our project. We can ask the it to generate a C4 diagram to show the high-level system overview. This diagram provides a bird's-eye view of the major components and their interactions:

• Frontend: The user interface where users upload their documents, sending them to the backend.
• Backend: The core logic that handles document upload, chunking, embeddings, storing them into the database, and also handles chunk searching when answering user's questions.
• Authentication Service: Manages user authentication and authorization.
• Database: Stores user's documents securely.

Step 3: Deep Dive into Components​

With the overview in place, we can zoom in on specific components for a more detailed understanding. Let's ask the AI to focus on the backend:

Now we can see that, in the backend, there should be a few components that will interact with each other. For example, for the document uploading and storing flow, we have:

• The Upload Service: to handle user's upload requests and send the document to the Embedding Service.
• The Embedding Service: to handle chunking and generate embeddings for each chunk before storing them to the database.
• The Vector Store Service: to handle chunk storing and retrieval from the database when needed.

For conversation, or user's Q&A flow, we have a few more services:

• The Query Processor Service: to handle user's queries, which will call the Vector Store to search for relevant chunks and send the chunks to the Response Generator.
• The Response Generator Service: which will communicate with an external LLM provider to generate the answer and stream it back to the Frontend.

Step 4: Security Considerations​

Security is an important aspect for this app, especially when we are dealing with user's private documents. Let's see if ChatUML can generate diagrams that highlight security measures.

Instead of the backend, we will dive into the details of the Authentication service and the Database. Ask the AI assistant to do so:

Interesting, so AI brings up Row Level Security, a PostgreSQL built-in security feature that makes sure a user can only access the database rows that belong to them. Great! We can use this for our application!

Step 5: Sequence Diagrams for Flow Visualization​

At this point, it seems like we have done enough exploring and have a good understanding of the system we are building. Before actually doing the coding, we can use AI to generate sequence diagrams to capture the flow of interactions within the system.

By leveraging the power of ChatUML, we've gone from a vague idea to a well-defined architecture for our "Chat to Document" application. This approach allows for rapid prototyping, efficient collaboration, and a more robust final product.

Now, are you ready to build your next product? Let ChatUML give you a hand, will you? 😁

We are updating the default model for all Free tier users from GPT-4o Mini to GPT-4o.

While GPT-4o Mini is cost-effective and fast, it has limitations, including a higher hallucination rate, making it less reliable for technical tasks.

By switching to GPT-4o, users will benefit from significantly improved generation quality, enhancing overall experience and usability.

For those who prefer or require GPT-4o Mini, it remains available as an option, especially useful when conserving credits.

Additionally, we are reducing the credit cost for GPT-4o and Claude 3.5 Sonnet from 5 credits per request to 3 credits per request, making these models more accessible.

Another big news for the month!

💥 Say Hello to Our New AI Friends​

We're super excited to welcome some incredible AI models to the ChatUML family:

• Claude 3.5 Sonnet: This coding whiz is the same brilliant mind behind tools like Cursor and Aider. If you love coding, you're going to have a blast with Sonnet!
• Perplexity Sonar 70B Online: Think of it as your personal research buddy that stays up-to-date with the latest internet data. Pretty cool, right?
• Gemini Pro and Flash: Google's smartest models have joined the party, and trust us, they're bringing some serious skills to the table!
• Qwen2.5 Coder 32B: Another rising star, this new Open Source model has gained alot of attentions for its coding ability!

Here's the best part - we've updated our pricing so you can play with any of these models as long as you have enough credits!

✨ Chat with Different Models Like You're Switching Between Friends​

Want to try a different model mid-conversation? No problem! Just click on the model name in your chat box and pick a new one. It's as easy as changing channels on TV!

We've also added a neat little feature that shows you how many messages you have left based on your credits. Think of it like a fuel gauge - when you're running low, you can switch to a more budget-friendly model to keep the conversation flowing. No surprises, just smooth sailing!

Keep chatting and creating amazing things! 💫

Big news! ChatUML just got a major update, bringing exciting new features and a fresh look to boost your productivity.

🦜 Floating Chat Box, More Space​

We’ve redesigned the interface with a floating chat box that you can move around freely, giving you more space to work on your diagrams without interruptions.

🖼️ Prefer Sketching? Just Paste an Image!​

Sometimes the best ideas come from brainstorming sessions around a whiteboard. Don’t erase your sketches when you’re done—snap a photo and paste it directly into ChatUML! The AI can turn your whiteboard notes into C4 diagrams or any other type of diagram you need.

You can now paste up to three images into the chat box and let AI work its magic!

Please keep in mind that all the attached images will only be used as a reference for the chat message you send. We will not store them on our server or display them in the conversation.

🌎 AI Assistant Can Access Web Links​

Have content stored elsewhere but don’t want to paste it all into the chat? Now, you can simply share a link! ChatUML will pull in the content from the link so you can easily reference it in your conversation.

Currently, we support text-based content, and PDF support is in the works. We’ll have it ready for you soon!

We are excited to announce that starting today, all users, including FREE users, will have access to GPT-4o Mini, the new and powerful model from OpenAI that outperforms GPT-4.

What This Means for You:

• Improved Performance: Enjoy faster and more accurate diagram creation and editing, thanks to the advanced capabilities of GPT-4o Mini.
• Enhanced User Experience: With the superior understanding and processing power of GPT-4o Mini, ChatUML becomes even more intuitive and user-friendly.
• Increased Accessibility: Now, every user, whether on a free or paid plan, can leverage the full potential of this cutting-edge AI technology.

Upgrade your diagramming experience with ChatUML and GPT-4o Mini today. Dive into the future of intelligent diagram editing and see the difference for yourself!

We have a quick and exciting announcement!

Starting today, we will be replacing the GPT-4-Turbo model with the new GPT-4o model. This is the most advanced model, with a whoppingly large context window of 128k tokens and stronger reasoning ability.

The new model is now available for all users.

Also, for users who purchased the Pro Package, we are increasing your message limit from 50 to 500 messages per diagram!

It's been a while since the last update. Over the past few months, we've been working hard to improve the product and the experience for our users and constantly release new updates, but it's time for an update post.

🚀 New models and faster response​

We've been rewriting the whole chat streaming backend to make it faster and more stable.

Our model selection has been updated with new models, including GPT-3.5 Turbo 16k and GPT-4. Here's the full list of models we're supporting:

• GPT-3.5 Turbo 16k: The default model for all users is now GPT-3.5 Turbo 16k, which supports 16k tokens, replacing the old 4k tokens model. That mean bigger and more complex diagrams.
• GPT-4: The most capable model, very good at logical reasoning and creativity. This model has the context window of 8k tokens.
• GPT-4 Turbo 128k: GPT-4 with the context window of 128k tokens. This model is still in Preview mode, so it may not be stable, and will be rate limited.

📖 New documentation site​

We've also released a new documentation site at docs.chatuml.com. This site will be the home for all the tutorials and guides for ChatUML.

🤫 One more thing...​

A few months back, we celebrated our 5000th users. As of today, we just surpassed 150,000 users 🎉. We're so happy to see the community growing and we're so grateful for all the support from our users.

As a thank you, we have a little gift for all users. Use the code FRIENDS150 to get 30% off when purchasing any package. This code is valid until 11:59 PM Feb 29, 2024 (PST), and can only be applied once per user.

First of all, I want to thank everyone for your love and support of our product.

When we opened up GPT-4 access to all users, everyone jumped in and tried it out. The usage during the past week has skyrocketed. There is no word that can describe our joy and appreciation to all the users.

Of course, the operation bill also skyrocketed. As a small team, this is a huge cost for us to handle, so we had to make a hard decision to limit access to GPT-4 model.

The GPT-4 model will only be accessible to users with 100 credits and up.

If you've selected GPT-4, but your current credit is lower than 100, it will automatically fall back to the GPT-3.5 Turbo 16k.

This change will not affect users who already purchased the Unlimited package. You can still access GPT-4 models as usual.

While we understand that any price adjustment might cause concern, we assure you that this decision was well-considered and necessary for us to maintain the exceptional level of service you have come to expect from us. The new pricing structure will enable us to continuously improve our offerings, enhance customer support, and ensure your experience remains exceptional.

Please don't hesitate to reach out if you have any questions or concerns.

Best regards,

ChatUML Team

Hello August! 🍁

Today, we landed the biggest update ever for ChatUML!

📝 Markdown support in chat​

From now, all chat messages will be formatted using Markdown, this makes it easier to read and follow what the AI assistant says, and you can also write code in the chat easily.

🔥 Streaming response​

We should do this from the beginning, but we did not! That was a mistake. Users had to wait for the full message to be received before seeing it on the UI.

From now on, you can start seeing the AI response immediately in the chat. This makes working with AI faster and feels more natural.

🦜 Better conversation context​

Previously, you may notice that only the context of the last message was preserved in the chat, so if you ask AI something from the beginning and a few messages later, the AI assistant will completely forget about it.

Well, that's no more. Now the full chat history will be preserved during a conversation.

This feature has a downside, though. You will reach the token limit faster! In that case, you can switch to another model, like GPT-3.5 16k.

We're happy to announce that ChatUML now supports GPT-4, the most powerful and capable model provided by OpenAI.

GPT-4 support comes with better logical reasoning, which means you can generate more high-quality and sophisticated diagrams.

You can also select the different models on the Settings page, currently, we're supporting 3 models:

• GPT-3.5 Turbo: The original one. Costs 1 star per request.
• GPT-3.5 Turbo 16k: Same as above but supports 16k tokens, that mean bigger and more complex diagrams. Costs 2 stars per request.
• GPT-4: The most capable model, very good at logical reasoning and creativity. Costs 4 stars per request.