The Alltius platform is engineered around a hybrid AI approach that fuses the adaptive capabilities of generative AI with the precision and determinism of symbolic AI. In an industry where every decision carries substantial financial risk, relying solely on probabilistic models such as large language models (LLMs) is insufficient. While LLMs are excellent at understanding and generating natural language, their probabilistic nature can lead to inaccuracies and hallucinations that are unacceptable in mission-critical applications.
Central to our solution is the symbolic AI layer, which comprises two complementary components: the KNO(ledge)-store and the (Work)FLOW- engine. These components work in tandem to ensure that every query or task is executed with absolute accuracy, supported by a robust and multi-layered data retrieval system that no single mechanism could achieve alone.
The KNO-store serves as the data ingestion, representation, management, and retrieval hub for the Alltius platform. It integrates multiple storage and retrieval mechanisms to form a cohesive system that leverages the strengths of each component:
Unstructured data is first transformed into a hierarchical representation that maintains contextual relationships. Advanced entity extraction techniques then identify key elements within the data, segmenting them into coherent chunks.
The segmented chunks are stored in a vector database that supports semantic similarity searches. This allows the system to quickly locate contextually relevant information based on the meaning of the query rather than relying solely on keyword matching. The vector database excels in capturing nuances and variations in language, ensuring that subtle but important details are not overlooked.
Complementing the vector database, our industry-specific knowledge graph maps the complex relationships between entities. For example, it can easily resolve queries such as “What are all the relevant policies for a 65-year-old customer in California?” by linking demographic data with policy specifics. This explicit representation of relationships allows for sophisticated multi-attribute lookups that surpass the capabilities of standard retrieval-augmented generation (RAG) systems.
Traditional structured databases and backend APIs are seamlessly integrated into the KNO-store. This ensures that both real-time and historical data are available and that transactional integrity is maintained.
To further enhance data retrieval across diverse sources, Alltius employs skills like text2sql and text2cipher. The text2sql skill dynamically converts natural language queries into SQL commands, enabling precise and flexible querying of structured databases. Meanwhile, text2cipher translates textual instructions into secure, encrypted queries, ensuring that data retrieved from sensitive sources remains protected. These skills work in a cascading fashion—triggering multiple, parallel retrievals from the vector database, knowledge graph, and structured databases—to consolidate a comprehensive, contextually relevant dataset for each query. No single mechanism could achieve this level of nuanced retrieval on its own.
By uniting these diverse data stores and advanced retrieval skills, the KNO-store provides a solid, contextually rich foundation. This ensures that when generative AI is called upon, it is grounded in accurate, verified information, effectively eliminating hallucinations and ensuring determinism.
Building on the robust data foundation of the KNO-store, the (Work)FLOW-enginemis tasked with interpreting user intents and orchestrating the appropriate sequence of actions. This engine functions as a state machine guided by an advanced intent manager, ensuring every step is executed with precision:
The FLOW-engine begins by analyzing the user’s natural language input or the output from the generative model. The intent manager categorizes the query, determines the requisite actions, and selects the appropriate predefined workflow.
Leveraging a deterministic state machine, the FLOW-engine navigates through each workflow step with step-by-step precision. This is crucial in the financial domain, where each step must be executed correctly to ensure compliance and reliability.
Incorporating Robotic Process Automation (RPA) technologies, the FLOW-engine ensures that each workflow is executed with minimal risk of error. This deterministic backbone is essential for handling mission-critical tasks where any deviation could have significant repercussions.
At strategic decision points, the FLOW-engine incorporates generative AI to introduce flexibility and adaptability. This means that while the system adheres to deterministic paths, it can adjust in real time to accommodate complex or unforeseen user requests—provided that these adaptations are always grounded in the verified data provided by the KNO-store.
The combination of deterministic orchestration and adaptable generative inputs creates a system that not only meets the high standards of precision demanded by financial services but also offers the agility needed to handle evolving customer needs.
Conventional generative AI systems typically rely on a single mode of data retrieval or execution, leading to several critical drawbacks:
Purely probabilistic models can fabricate details in ambiguous contexts. By anchoring generative responses in a meticulously curated symbolic foundation, Alltius virtually eliminates such hallucinations.
Without a multi-layered, symbolic approach, generative systems can produce inconsistent outcomes. The deterministic state machine of the FLOW-engine, combined with the rich data interplays within the KNO- store, ensures that every outcome is both accurate and repeatable.
Standard RAG systems fall short when queries demand nuanced, relationship-based lookups. Our integrated knowledge graph, in collaboration with the vector database and structured databases, enables multi-dimensional data retrieval that no single system could achieve on its own.
By incorporating text2sql and text2cipher, Alltius facilitates the simultaneous retrieval of data from multiple sources. This cascading approach synthesizes the strengths of various retrieval mechanisms to deliver a holistic, precise answer—ensuring that every query is resolved with depth and security.
Alltius’s symbolic AI layers—comprising the KNO-store and the (Work)FLOW- engine—establish a new benchmark in AI-driven financial services. By integrating advanced retrieval mechanisms that interoperate seamlessly across vector databases, knowledge graphs, and structured data sources, and by leveraging cascading skills like text2sql and text2cipher, our system ensures that every generative output is grounded in accurate, context-rich data. This robust, hybrid approach not only mitigates the risks associated with conventional generative models but also delivers a dynamic, reliable solution for mission-critical financial workflows, making Alltius the superior choice for the industry.
