You can’t swing a bat without hitting something AI these days.
TrillaBit has also stepped into this realm, and we’re genuinely excited about it. Our aim is to provide you with a solid foundation you can rely on.
We get it, cutting through the noise to see what’s really going on can be tough. Somewhere between the lofty predictions of the future and the more tangible implementations like chatGPT (along with numerous other LLMs and RAG pipelines), there is solid ground to stand on.
AI encompasses a wide range of algorithms, techniques, and applications. When you mix this already complex tech world with people’s imaginations, well… understanding the current state can be a bit fuzzy. Often, what you see is a prototype representing just 20% of a vision, followed by a lot of talk about the vision.
I asked chatGPT why the term ‘AI’ is confusing, and here’s the gist of the response:
"Overall, the term 'AI' can be confusing because it represents a diverse and evolving field with implications that extend across technology, ethics, society, and more. Clarity often comes from breaking down AI into its components, understanding its current capabilities, and discussing its potential impacts in specific contexts." ~ChatGPT
Given the confusion around the current state, let’s start by clearing some of this up. Describing the current state is tricky because things are evolving rapidly. It’s a bit like driving a car—you don’t look out the side windows to figure out where you are. When you're going fast, you focus on the road ahead to see where you're headed and those signs just up ahead to know where you are. You might not see the entire path, but you have a good sense of where you're headed.
GenAI Today
A scope limited perspective
If you ask ChatGPT what applications AI covers today, it will provide a long list and then tell you, these are just a ‘few things’. Obviously the impact on our lives is quite extensive.
What I’m seeing in our space today however still has its limitations. For instance, LLMs are amazing at natural language processing. They have been trained across vast amounts of text. But even so, we’ve still had to augment models with search engine, vector database technology to now include proprietary data within enterprises. RAG pipelines are one way to build a model from some static data and enhance it with other text on implementation, typically proprietary data.
Then we have NLP(Natural Language Processing) for SQL. It’s more task-specific than the general-purpose LLMs like GPT. The primary focus of NLP for SQL systems is to understand and convert natural language queries into SQL, making it a specialized subset within the broader field of natural language processing. But NLP for SQL still isn’t able to create highly complex SQL statements yet, and often still makes mistakes. There’s still a lot of work to do here… If you’ve seen even a portion of the data models I have in my career, you’ll know the craziness that people have implemented. From the strangest naming conventions, to the ugliest of models…. Let’s just say 3rd normal form and above have certainly not been respected almost anywhere. Every implementation is different and learning from this certainly isn’t generic. Although I don’t doubt that AI will be able to navigate this someday, I think we’re a ways off. Today, it’s one model at a time.
Databases are here to stay, and so is the need for data security. We not only have to deal with proprietary data models and information, but also the challenge of securing this data across different roles, locations, and organizational levels.
I believe this is why, for the past year or so, AI in the analytic tooling space has been tied to either tools for developers themselves, or they’ve been applied to a limited context, like spreadsheets. If you have access to a spreadsheet, then you have access to all the data within that spreadsheet, and therefore we don’t need to worry about security when using nlp to dynamically build some dashboard on the spreadsheet.
For developer tools, it’s an efficient assistant. Let AI build out the code quickly, then the developer can enhance it (fix it) and implement it without the security risk.
Currently the best implementations today involve a human in the middle or simplified context. Nothing wrong with this, but why does it need to be a developer in the middle?
Bucking the Trend
TrillaBit is aiming to buck the current trend. Our vision isn’t to build tools for developers, but build out self-service to ALL users while accommodating the AI inconsistencies that exist today. Practical implementations that users find useful and can work with.
Let me explain... The bandwagon trend for many tools today is to implement whatever exists in AI so they can give it to developers. Let’s say NLP for SQL. As a developer you need to build out dashboards, so you use an AI plugin to ask a question and it will return a SQL statement for you. Now you can use that SQL statement to build out your visualization. Although you may need to tweak it, because it didn’t really translate your ambiguous language well enough and it can’t really create that more complex query that you actually wanted, but hey! you saved a little time, and the product gets to say they’ve implemented AI! Win win!... sort of.
Bucking the Trend: What if… instead, AI didn’t return SQL? What if instead it returned something more simple. Like tags, tags that are not ambiguous and relate to more complex sql statements. Something that is easy to understand, and modify to get exactly what you want!
Imagine asking a question and getting back a graph visualization, and a set of tags that you can easily modify because they’re not as complex as the SQL language. All data returned is secured to you and you can now drill down into that visualization to explore your data.
The SQL is still generated in the backend and securely executed against your data model, but presented in an easy to use way. A more user-friendly way.
TrillaBit GenAI Features
When it comes to new industry and life changing tech like GenAI, we need to push the envelope and reach for the real value. Copying the next guy might give companies something to talk about, but TrillaBit strives to deliver value. To deliver practical GenAI features directly to the end user.
TrillaBit is already built to leverage metadata over traditional development tools. This approach allows for dynamic control over data exploration, analytics and collaboration with embedded security. Enhancing this dynamic platform with GenAI just makes sense and is the obvious next step in our evolution.
I’ll just highlight a few of the features here, as what we’re actually doing would take so much more time to get into. And we will, so stay tuned!
Early TrillaBit GenAI Features:
Natural Language to a fully visualized, editable, drillable, sharable and secure KPI. Far beyond a SQL statement.
Most end users don’t want to have to deal with technical languages. They want to get on with running their business. Allowing for Natural Language gets us one step closer to this.
If you want to understand more about TrillaBit’s more robust solutioning, please feel free to reach out!
By training our models on thousands of datasets, AI can start to learn which attributes you are using within your context. With a simple explanation of the dataset, we can feed our AI the meta data to provide descriptions of each attribute within.
This removes a great deal of tedious work we don’t really wish on anyone. Ask your database people, in many cases, this level of valuable documentation just isn’t done…. Because no one wants to do it.
Our Gen AI can also provide descriptive text on what the resulting KPI is. This is useful for end users who might not fully understand what the actual analytic is just based on a title and some axis labels.
Just like other levels, this often isn’t done. But then the analytics are delivered and people are asking… what is this again? Again, this level of valuable documentation just isn’t done, because no one wants to spend the time doing it.
This is nothing short of amazing! Based on the metadata and the context of a dataset, Our AI can automatically create a basic dashboard for you with many kpis specific to the context. And all the descriptions added for you.
Going beyond the norm, because we are not just generating SQL, all of the Security and functionality will already be built into each KPI within the dashboard. Drill down, Modification, Sharing, ability to quickly edit the KPI and copy it to other dashboards. All for end-user self-service. Amazing!
As is today, dashboards can remain domain, workspace, group or user level. If you want to learn more about the TrillaBit platform and the amazing things we’re doing already, please contact us!
When you create a kpi, you’re essentially answering a business question. But what if the system can provide alternative perspectives on that question?
When you create a KPI, you will receive a list of possible alternative KPIs around the same context as your business question. Essentially providing you with different perspectives.
When you work with a mentor or colleague, they may often provide you with feedback or alternative thinking as you work through a problem. This is a type of co-pilot implementation to help you on your data exploration journey. Courtesy of TrillaBit and GenAI!
We have a lot… but we’re keeping a lid on that for now.
In summary I believe TrillaBit is making great strides in innovation and providing a platform to benefit B2B SaaS companies as they strive for competitive advantage in this newly fast moving GenAI world.
We will expand on this as we go, but if you would like to know more, sooner, Please feel free to reach out to us! We’ll be happy to get into more details!
Thanks,
Keith