How We Lead: AI Technology – expert.ai


In the evolving world of AI-driven business solutions, being recognized as a Leader is a significant achievement. For us, expert.ai’s inclusion in “The Forrester Wave™: Document Mining and Analytics Platform, Q2 2024” as a Leader underscores our commitment to excellence and innovation. We believe receiving the highest scores possible in the Knowledge-based or Symbolic AI, Generative AI (GenAI), GenAI Pre-processing, GenAI Post-processing and Document Labeling for Machine Learning (ML) training criteria highlight the strengths that set us apart. Let’s explore how expert.ai excels in each of these critical areas.

Knowledge-based or Symbolic AI: The Core of Hybrid AI

At the heart of expert.ai’s technology is our integration of the best of Knowledge-based or Symbolic AI for Hybrid AI, which is the combination of machine learning (ML), symbolic AI and large language models (LLMs). This combination lets us automate complex business processes and harvest relevant knowledge and actionable insights from all your organizational data. Our technology is based on a rich set of domain-independent representation of knowledge and industry specific taxonomies and ontologies. Our analytic capabilities include deep linguistic analysis, key phrase extracts, named entity recognition (NER), relation extracts and sentiment analysis.

The deep linguistic analysis combines text subdivision, part of speech (POS) tagging, morphology, lemmatization, syntactic and semantic (or disambiguation). We resolve ambiguity by performing disambiguation using our embedded semantic engine and Knowledge Graph. With millions of concepts out of the box, our users can effortlessly customize the platform to add domain- and enterprise-specific knowledge and proprietary machine learning (ML) and deep learning algorithms without vendor support.

GenAI: Versatile and Controlled AI Applications

Expert.ai’s approach to GenAI offers unparalleled flexibility and control. Our platform enables both full and assisted control over LLM prompts, ensuring that these models are used purposefully rather than as unchecked “oracles.” Our LLM tasks include:

  • Zero- or few-shot extraction of single or multiple data points
  • Custom summarization of texts (including proper summarization, schematic abstracts, bullet lists, datapoint lists, JSON dictionaries, etc.)
  • Q&A tasks providing questions and text fragments with candidate answers to generate optimal answers using retrieval augmented generation (RAG)

Whether it’s extracting data points, summarizing text or answering complex questions, our model-agnostic solution allows for seamless integration of any API-based model. The curated prompt library and efficient production processes, like document segmentation and parallelized LLM calls, ensure that our customers experience optimal performance with minimal latency.

GenAI Pre-processing: Enhancing Data Preparation with Advanced Tools

Pre-processing is a crucial step in any AI workflow, and expert.ai excels in this area by providing tools that significantly streamline data preparation. Our platform can quickly create symbolic models for pre-annotation of complex datasets, leveraging active learning to speed up and semi-automate document annotation with dynamic learning.

Integrated with GPT and our proprietary LLMs, the platform can be seamlessly integrated with symbolic and ML models and workflows. The same models are also usable in an advanced RAG-based architecture specifically oriented at question & answering (Q&A) tasks over complex document collections (e.g. manuals, claims, contracts, structured PDFs, etc.). An out of the box prompt library includes summarization and extraction prompts.

The platform supports document splitting to improve LLM efficiency, control prompt input and address context size limitations and supports various context sizes, from 4K to 32K, based on the model. Workflows automatically segment documents to maximum context size and perform multiple LLM calls to get responses for a downstream reconciliation and harmonization of the obtained results.

GenAI Post-processing: Maximum Control for Superior Results

The expert.ai platform offers unparalleled control over LLM behavior in document post-processing workflows, enabling precise document segmentation and cleaning. This ensures that LLMs work within well-defined scopes, reducing the risk of hallucinations, erroneous outputs or undesired results.

Moreover, the platform provides traceability from LLM-derived outputs back to their original document sources, enhancing traceability, transparency and accountability. Custom nodes and custom linguistic symbolic rules can be utilized to further validate the content and structure of LLM outputs, ensuring that only the highest quality data is produced.

Document Labeling for ML Training: Precision and Efficiency

Document labeling is a vital part of ML training, and expert.ai’s platform is designed to make this process as efficient and accurate as possible. Our multilingual semantic engine fully disambiguates each term in the text, while the symbolic rule-based IDE allows for precise extraction and labeling of custom data fields. The platform also supports thesaurus management.

The output of the text analysis from the custom linguistic model can be used to pre-annotate the training and test set to simplify the annotation process. Active learning is available to dramatically speed up the annotation process.

With access to 25 industry- and domain-specific ontologies (including finance, insurance, publishing, intelligence, medical records and emotions) and support for leading standards like Geo names, MeSH, Wikidata, SNOMED-CT, ICD 10, UMLS and SKOS, our users can tailor the knowledge graph to their specific needs. LLM-based document processing workflows further enhance this process by creating labeled datasets that integrate seamlessly into our authoring environment.

Conclusion

Expert.ai’s leadership in the Document Mining and Analytics space is driven by our strengths in Knowledge-based and Symbolic AI, GenAI and advanced data processing techniques. As we continue to innovate and refine our platform, we remain committed to providing our customers with the tools they need to unlock the full potential of their data. We believe our recognition as a Leader in the Forrester Wave report is a testament to the value we deliver and the impact we make across industries.

Learn more about How We Lead: AI for Regulated Industries and Strategy, Vision and Innovation.



Source link