Wrapping up 2024

In 2024 we implemented hundreds of features, made our platform 10x more stable and secure - In addition to significantly growing from a business perspective. If you look at what we were able to implement, ship, and deliver in 2024, the list might seem daunting for most. For us at AINIRO it was just another year where we were 10x as productive as most others in our field. Below are the highlights of what we were able to wrap up in the year that passed.

AI Agents

In 2024 we released a stable implementation for having the LLM invoke AI functions. This allowed us to integrate our AI technology with your existing databases, APIs, and literally anything you can imagine. This turned our platform from a passive text generating LLM based upon RAG, to a completely horizontally integratable AI Agent foundation.

In addition, we integrated the meta-programming scaffolding process into our AI function implementation, allowing ut to use meta prompt engineering to automatically build AI agents based upon Low-Code and No-Code. This was only possible because of Magic Cloud, being a No-Code and Low-Code platform, allowing us to build incrementally and exponentially, one layer on top of the other.

CRUD generator

We significantly improved upon our backend CRUD generator, most importantly allowing us to generate aggregate endpoints, distinct endpoints, and group by endpoints, having the LLM generate pivot tables, and grouping information to aggregate data and information from connected databases.

In addition we significantly simplified the meta prompt engineering occurring during scaffolding, resulting in much higher quality matching towards the RAG database as users are requesting function invocations.

We also integrated these features into our AI Agent infrastructure, allowing us to incrementally wrap your database into an AI Agent, to allow for CRUD and business logic integration with existing databases.

Combined with our ability to automatically import CSV files, and having a structured SQL database generated from its columns - This resulted in interesting features such as the ability to use natural language to interface with your CSV files.

As a final feature to lists here, I would like to mention the ability to use meta prompt engineering constructs to generate AI functions from Hyperlambda files, either individual files, or by recursively traverse entire folder structures with a simple button click.

AI Expert System

In 2024 we also resurrected our AI Expert System, and significantly improved upon it. The AI Expert System is now easily deployed from your plugins component, allowing you to rapidly get it up running.

The AI Expert System also supports SSO through OpenID Connect, allowing users to authenticate using their existing Google accounts, and other OIDC providers. We also significantly improved upon the system's graphical user interface, in particular for small devices, such as iPhone phones and Android Phones. An example here is the ability to render tables correctly, by adding horizontal scrolling if the table overflows the available width.

We also optionally added the ability to share conversations you're having with the LLM, through a share button in the system. The idea being that this can create a windmill SEO effect, where users can share their conversations in social media, and such help onboard new users through simply using the system and sharing results with others.

In addition to the above, we also implemented support for Whisper API and TTS, allowing you to use voice commands to give the system commands, and have it respond with a natural sounding voice to your requests.

AI SaaS in a box

In 2024 we helped launch several AI SaaS apps for clients. This was based upon our ability to white label the AI Expert System and connecting it to external datasources through our AI Agent technology. Probably the most prominent SaaS apps here are the following you can see as examples of our technology.

We're now almost at the point where we can implement, deploy, and deliver an AI Agent-based SaaS app in some few days.

Embeddable AI chatbots

We also completely rewrote the embeddable AI chatbot's graphical user interface (GUI). Some of the things we added were 100x stronger CSS, not interfering with your website's existing CSS in any ways - In addition to adding a popup parameter, that you can see on our website. The embeddable AI chatbot basically had a complete make over on its GUI.

In addition we added support for associating animations with the chatbot button, making it more prominent and visible on your page, further increasing conversions and engagement.

Security and stability

In 2024 we added the ability to automatically block users abusing your AI chatbot. This is useful if some users are spamming the LLM with lots of irrelevant questions. In addition we fixed a lot of other issues related to how we create embeddings for RAG data, significantly improving the platform's stability.

We also implemented our own PoW-based CAPTCHA library, significantly reducing the bandwidth usage for an embedded chatbot, resulting in much faster load times and less latency in our tech.

Misc

In addition to all of the above things, we added the ability to have our AI technology automatically generate charts and visual representations. This was particularly important for Fintell.AI to have it generate visual charts based upon financial information.

Our technology can now also trigger the embeddable AI chatbot based upon a query parameter, allowing you to send "hyperlinks to conversation starters", to increase engagement and automatically initiate discussions with the AI chatbot.

You can now also import images directly into your machine learning type, allowing you to mass upload images, and have the AI chatbot or agent showing these based upon VSS matching.

We also added the ability to have AI chatbots and agents with "long term memory", allowing users to create RAG training snippets, during a natural conversation with the LLM.

Allowing users to upload a "URL list" as a CSV file, which results in the AI chatbot crawling and scraping all of these URLs, generating training data in the process.

AI functions

Below is a list of some of the AI functions we implemented.

Send email
Generate image using DALL-E
Keyword search through RAG data
Anything resulting from using the backend generator to generate CRUD endpoints
100+ additional AI functions

Optimisations

In addition we updated the speed and performance of the backend by 10x during 2024. In addition to upgrading to .Net 9, which by itself improves the speed of our platform a lot, we significantly optimised the way we retrieve RAG database items from the database, resulting in significant memory and CPU improvements, allowing us to deliver larger and faster AI chatbots and AI agents.

Wrapping up 2024

Since we started using a changelog the 22nd of May in 2024, we've got 185 changes to our system, and 69 new releases. The above isn't anywhere close to the full list of changes we did in 2024, but only partial in fact. In 2024 we had 706 commits to Magic's GitHub repository, and we've got 50+ additional repos. So we definitely weren't lazy.

We suspect 2025 will be an equally productive year, and we wish all of our clients and partners blessings in 2025, and may all your dreams come through.

Have a Custom AI Solution

At AINIRO we specialise in delivering custom AI solutions and AI chatbots with AI agent features. If you want to talk to us about how we can help you implement your next custom AI solution, you can reach out to us below.