Our Team

Our team comprises experienced researchers and developers with expertise in graph algorithms, knowledge representation, NLP, and distributed systems. We are adept at designing and implementing decentralized, fault-tolerant architectures and are committed to open-source collaboration. Our collective background includes significant work in LLM-RAG AI and large-scale data processing, with a strong mandate for ethical AI development.

Company Name (if applicable)

Project details

The pursuit of Artificial General Intelligence (AGI) necessitates robust, scalable, and ethically aligned knowledge management systems. However, the dynamic and often noisy nature of real-world information, particularly in domains such as scientific discovery, open-ended learning, and evolving human knowledge, presents significant challenges to the construction and maintenance of high-quality knowledge graphs. Current centralized or simplistic cloud-based architectures often struggle with scalability, resilience, and the integration of nuanced, ethically informed reasoning.

The Orpherium framework will deliver advanced tools for interfacing with, augmenting, and evaluating these distributed knowledge graphs, specifically focusing on refining noisy data, detecting inconsistencies, and supporting ethical, impartial reasoning within neuro-symbolic AI contexts, with good compatibility or direct utilisation of MeTTa and MORK. The approach we are advocating for builds resilience, horizontal scalability, and distributed intelligence, aligning with Hyperon's decentralized infrastructure and the promise of neuro-symbolic AI.

Orpherium emphasizes the full lifecycle of knowledge graphs, from the initial ingestion of raw data to its continuous refinement, validation, and ethical application:

• Creating and Populating (Aggregators): The lowest level, Aggregator nodes, are responsible for autonomously extracting raw or semi-structured data from diverse sources (e.g., IoT device streams, scientific publications, raw text, JSON, MeTTa expressions). They perform initial parsing and structure this raw input into nascent, distributed graph fragments.

• Refining and Distilling with Network Sharding (Coherents): Coherent nodes receive these initial graph fragments from Aggregators. They perform the crucial task of integrating and distilling the information, detecting redundancies, resolving basic inconsistencies, and applying schema-agnostic structuring to form more compact and semantically meaningful representations. Critically, Coherents are responsible for network sharding of these knowledge fragments across the massive P2P storage and processing network. This means they intelligently break down and distribute portions of the knowledge graph across various nodes, ensuring efficient storage, retrieval, and parallel processing. This sharding mechanism is vital for horizontal scalability, fault tolerance, and enabling the system to operate on truly massive, evolving datasets without any central bottleneck.

• Detecting Inconsistencies and Proposing Resolutions (Oracles & Human Trust): Oracle nodes operate at a higher level, receiving refined knowledge from Coherents. They perform sophisticated consistency checks, identifying contradictions, invalid patterns, or anomalies within the evolving knowledge graph. These Oracles are specifically trained on empathic principles, ethics, and aesthetics, ensuring their judgments are not just logical but also aligned with beneficial human and collective values. Their "empathic and impartial" nature is crucial for fostering trust. When Oracles identify areas where the knowledge graph's ontology or algorithmic variance needs adjustment – for instance, to resolve a complex ethical dilemma or incorporate a new paradigm they don't simply make the change. Instead, they formulate these insights as proposals.

• Fact-Checking and Grounding (Oracles): Oracles are designed to fact-check external outputs (from LLMs or other AGI components) by querying and cross-referencing information within the refined knowledge graph, ensuring grounded and accurate responses.

• Bidirectional Knowledge Feedback & Human Governance (Oracles and Sage Oracles to Lower Levels): A key differentiator of Orpherium is its bidirectional knowledge flow. While the "upward wave" refines raw data into higher-level insights, the Oracles and Sage Oracles also transmit refined knowledge, ethical guidelines, anomaly flags, schema updates, and priority adjustments back down to the Coherents and Aggregators. However, the modulation of the a knowledge graph's ontology and fundamental algorithmic variance (how it learns and interprets information) is subject to human oversight. Oracle-generated proposals for such critical changes are presented to the global network of decentralized node supporters. These supporters then vote on proposals using liquid democratic rules, allowing for a dynamic, informed, and transparent governance model. If approved, these democratically sanctioned modulations then become part of the "downward flow" of refined knowledge and guidance, ensuring that the AGI's foundational understanding evolves in a way that humans can understand and trust. This continuous feedback loop ensures that lower-level agents are constantly learning and adapting based on the higher-level intelligence and ethically-governed considerations of the system.

• Designing Interfaces for AGI Interaction (All Nodes, especially Sage Oracles): We will explore exploratory APIs and conceptual interfaces demonstrating how AGI systems can actively reason with, act upon, and update the Oracle Federation. This goes beyond simple querying, enabling AGI to "co-evolve" with its knowledge, with the Sage Oracle providing the highest-level interaction point for complex reasoning and advanced communication, always within the bounds of democratically aligned principles.

• Annotating and Weighting (All Nodes): Each node type will contribute to annotating entities and relations with confidence scores, semantic types, relationship types, and provenance information, providing richer context and enabling bias mitigation for downstream reasoning.

• Enabling Continuous Updates (Upward Wave & Hierarchical Sharing): The inherent "upward wave" of data reprocessing and the hierarchical data sharing amongst oracles naturally support continuous graph updates. Lower-level nodes (Aggregators, Coherents) feed new information to higher-level ones, enabling real-time obsolescence detection for outdated nodes/edges and dynamic propagation of new knowledge throughout the network.

• Evaluating Effectiveness: We will develop benchmarks to evaluate the effectiveness of the Oracle federation in supporting complex AGI reasoning tasks like multi-hop question answering, analogical retrieval, and hypothesis generation.

• MeTTa and MORK Integration: Outputs and internal representations will be designed for compatibility with MeTTa expressions and MORK. Our intention is for Oracles to directly utilize MORK and similar Rust based infrastructure as a hypergraph backend for real-time reasoning and retrieval over vast number of symbolic spaces, accelerating cognitive synergy within Hyperon-based agents.

The "upward wave" represents the continuous processing and refinement of data, moving from raw inputs at the Aggregator level, through coherence-building and sharding by Coherents, to sophisticated reasoning and ethical guidance by Oracles, culminating in cross-domain synthesis and advanced communication capabilities at the Sage Oracle level. The "downward flow" of bidirectional knowledge feedback ensures that insights from higher-level reasoning influence and refine the operations of lower-level agents, with critical modulations being subject to human approval. "Hierarchical data sharing amongst oracles" ensures that refined knowledge and emergent insights are distributed and utilized effectively across the network, driving adaptive and ethically aligned behavior across the entire AGI system.

Open Source Licensing

The entire Orpherium project, including all code, documentation, and tools developed, will be released under the MIT License. There are no exceptions; all components will be fully open-sourced under this permissive license, fostering community collaboration and transparency.

Background & Experience

Our team brings extensive experience in distributed systems, AI architecture, knowledge graph development, and ethical AI. Steve M. has many years of experience in graph database design, query optimization, and distributed graph processing, with specific contributions to Sun mircosystems and Oracle. Victor S specializes in knowledge representation formalisms and ontology engineering for complex, dynamic, and distributed systems. Thomas Cal has extensive experience designing and implementing decentralized systems, peer-to-peer networks, and fault-tolerant architectures, crucial for the distributed nature of the oracle federation and its network sharding mechanisms. Our team members have a basic understanding of MeTTa/MORK and the intended direction.  We have a strong track record of contributing to open-source projects and are comfortable with GitHub-based workflows.

Links and references

https://vimeo.com/1088105385?share=copy

more to be added

Additional videos

Describe the particulars.

various AGI summits online