Smabbler Language Engine Knowledge Graph

Smabbler Language Engine automatically transforms open text into highly precise structured knowledge for versatile applications.

It helps to generate real-time insights for advanced analytics and data-based decisions.

Problem within data/knowledge discovery in pharma companies:

Thousands of incoming patient/pharmacists/scientists inquiries are related to drug safety, usage and storage restrictions. Analysts look for symptom and context patterns within the incoming query stream. Because of domain and portfolio complexity critical data hidden in communication is overlooked.

Smabbler is a Cause-effect Language Reasoning Engine and Knowledge Graph with huge general knowledge representation.

Its key highlights:

• Ability to build automated ontology
• Knowledge generalization
• Learning instead of training
• 90%+ accuracy out-of-the-box
• Works with different languages
• Works with different data sources / formats (pdf, email, url, etc.)
• Domain and sector agnostic

At every level of a pharmaceutical company there are multiple departments all working on textual medical data.

All these mundane tasks - like contextual analysis and correlation of
incoming inquiries - require actionable knowledge: structured and
correlated data for analytical insight / trend identification.

Smabbler provides a cognitive layer on top of the company's data repositories and automated the process of knowledge discovery in textual data assets. These enhanced knowledge sets can greatly improve and automate multiple processes and applications in the organization:

• Analysts - identifying, classifying and structuring detailed information for analytical insight / trend identification
• Researchers - maintaining master data / ontology for drugs portfolio and clinical trials
• Customer Agents - using recommendations for answers to patients, doctors and pharmacists

Smabbler Language Engine performs analysis of communication data from patients, doctors, scientists and automatically structures in a transparent knowledge database for precise insight and market feedback.

Smabbler identifies information about diseases, the patient's health, side effects, disease symptoms, drug / drug-drug interaction, dosage, drug storage.

Newly discovered data can be used to:

• identify trends and new signals (e.g. new symptoms for patients, drug interactions, side effects of therapy) for risk assessment & management
• provide feedback to improve drug safety, dosage, and therapy as well as provide data for medical research.

The problem of natural language understanding is central as it is a prerequisite for many business process tasks, where text-to-knowledge extraction is one of the most important ones.

Current Machine Learning (and Deep Learning) models that utilize neural networks for language tasks, amount to statistical fitting and do not exhibit 'real' understanding of natural language.

The problem with lack of cause and effect analysis, and abusing of statistics in language processing is highlighted by Turing Award winner - Juda Pearl. This globally recognized pioneer of AI and Machine Learning now becomes one of its sharpest critics ('As much as I look into what’s being done with deep learning, I see they’re all stuck there on the level of associations. Curve fitting.').

How Smabbler solves it:

Smabbler hybrid technology bases on causal reasoning (cause-effect text analysis) combined with a knowledge network (multidimensional knowledge graph and ontology representation). This way the technology is able to understand context, infer from text, build automated taxonomies and ontologies, work with complex documentation, and easily handle various domain use cases.

Smabbler proprietary technology combines:

• Cause-effect language reasoning engine (NLPengine), a hybrid solution that processes complex language tasks utilizing self-learning approach and
• Knowledge Graph querying.Knowledge Graph database (KG), which is the framework for domain independent ontology representation created automatically from unstructured text corpuses.

We developed a unique domain-independent automatic ontology generation graph-based framework. It is a multi-relational and multi-dimensional knowledge graph structure able to handle concepts, definitions, syntax and their relations. We use feedback loops from language engine–graph communication to identify new knowledge and convert it into ontological form.

Open problem in AI:

Even in today's technology advancement, the biggest open problems in AI are related to natural language understanding. Until there are no systems that read and understand text the way a person does, all of market progress focuses merely on improving systems' ability to do pattern matching.

A general horizontal language engine is the first step toward achieving general artificial intelligence able to address complex human cognitive tasks, thereby paving the road to truly innovative solutions enhancing the lives of millions.

10 people core team (full time) 

(+ external technical support team (approx. 20 people))

• MIT EF acceleration program
• Microsoft Partnership Network program
• NVIDIA Inception program
• PwC Startup Collider program
• plus few global companies (clients)

1. MIT - possibility to cooperate with this brand is one of the biggest prizes there could be.
2. Recognition - is more than welcome for every startup.
3. Grant - it always helps in innovation development.

MIT - there are a lot of bright minds we really admire (from language / cognitive technologies / AI areas)

Access to the right knowledge is one of crucial factors in the healthcare area. From the very beginning our work is driven by the goal to serve people with quick access to needed knowledge.

We continuously work on our technology improvement. That's why the prize could help us extend our R&D team and accelerate further innovation.

From the very beginning we are focused on technology innovation and questioning 'the impossible' - making computers understand humans.

That's why we continuously explore new ideas and technology areas. We innovate by joining different science concepts (life-long learning, graphs and general ontology).

We develop general-purpose language technology (cognitive engine) able to work horizontally across multiple domains (we're domain agnostic).

The purpose is to provide the universal scalable solution - Language AI.