First True AI Slack App

Few days ago DataLingvo released what I would consider the first Slack app that delivers one of the true AI traits: a free-form natural language comprehension.

What is DataLingvo? DataLingvo is a Siri-like interface for your business data.

It supports Google Analytics and number of custom data sources like or SQL databases with more coming up soon. You can type or speak to it to ask any questions about your data using a normal human language — and it will provide back a comprehensive answer. You can chat with it using web browser or a growing number of supported messaging bots (with the Slack app being the first one out of the gate).

We at DataLingvo think of it as a first true Citizen Data Science platform.

DataLingvo technology has several unique characteristics that I’d like to highlight here.

Free-Form Natural Language Comprehension

Most, if not all, of the current systems with question-and-answer NLP-based capabilities are built on what is commonly known as ”NLP on rails”. These systems are essentially built around a few dozen of predefined words and phrases (with synonymical and grammatical variations). Some of them also understand some basic date ranges, geographical location, etc. Basically, these are a custom query languages technically masquerading as natural language but with massive restrictions (specific expected grammar, order of words, no typos, no extra words, no context, etc.). In these systems, step left or step right from these rules always leads to a failure to understand the user’s input.

History is littered with companies and academic projects building these types of systems with various degrees of sophistication. And historically users have soundly rejected these systems once they figured out that despite of claims of “natural” language — they still need to memorize all the special words, rules and restrictions to use these systems — and it turns out to be no easier than learning any other custom query language…

Another, more recent, problem with this approach is the fact that although written (typed) natural language queries tend to be more formalized, spelling and grammatically correct — a spoken language is dramatically less so. Not only current speech recognition is rather iffy beyond basic well-known words but we also tend to speak in much less grammatically correct (or ideal) way. All that makes “NLP on rails” very questionable in speech-oriented interfaces… which will be like 90% of them in the next 10–15 years.

At DataLingvo we’ve attempted to build a true free-form natural language comprehension system that will be good enough to answer data analytics questions and be free of the limitations of the typical “NLP on rails” systems.

Knowing that classic computational linguistics methods won’t solve this problem we developed a patent-pending Human Curated Linguistics (HCL) technology that is based on recent Semantic Grammar research and solves the “last mile” problem of language comprehension with real-time human curation plus sophisticated supervised Reinforcement Learning capabilities to self-learn more with every interaction.

All in all, we can understand and answer pretty much any question related to your data source in any form you want to ask it. For example, type this into your Slack (you need install the Slack app first):

or you can be as adventures as:

or you can be dead-on terse:

and you get same result in all cases:

DataLingvo Slack app (test data)

and via seamless hand off to our full featured web app you can get pretty sophisticated in your data analysis now that you have all your data at fingertips:

DataLingvo web app (test data)

You choose what and how to ask. No special words to remember, no rules or particular grammar to follow. Just a normal everyday language you already know and use.

Guaranteed Correctness and Determinism

Another important feature of DataLingvo technology is guaranteed correctness and determinism of the answer.

It’s still surprising for many newcomers that beyond very few “NLP on rails” systems today most of the general NLP/ML-based systems are non-deterministic, i.e. they can only provide an approximate or probabilistic answer. Scores of sentiment analytics systems, content comprehension, IBM Watson and Watson-wannabe systems do not guarantee the correctness of the answer (some of the time or all the time).

And while determinism or correctness isn’t a big deal for something like a sentiment analysis (few would care if their brand’s mentioning is 75% or 75.5% positive on a given date) — for business data analytics and reporting this is crucial. If I’m asking for the total revenue for the last quarter I’m expecting to get the right number to the penny.

DataLingvo is built from the ground up with determinism in mind. We either provide the answer and guarantee its correctness or we don’t (and we’ll tell you why). If you are asking a question that is impossible to answer given your current data source we can determine that and respond accordingly.

From the technical point of view our HCL technology doesn’t allow a non-curated and unsupervised results into our self-learning system thus guaranteeing that all — first-time curated and auto answers that we’ve learned to answer before — are correct.

For example, go ahead and ask this:

and you get predictable result (Google Analytics doesn’t know anything about moon’s temps):

DataLingvo Slack App (test data)

(but we are smart enough to suggest couple of links where you can get that answered very quickly).

Slack Deep Linking & Single Sign-On

One the coolest features of our Slack app is its full bi-directional integration between web app and Slack app. Slack app provides unique seamless hand off to the web app via Slack deep-linking and single sign-on. All the data is synchronized between all different client apps (i.e. Slack app, upcoming Telegram bot and web app) — inbox and pending answers, account & team information, data sources, examples, and dashboard:

Dashboard at DataLingvo Web App (test data)
Dashboard at DataLingvo Slack App (test data)

Take DataLingvo Slack app for a spin. Install it directly or find it in Slack app directory.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.