O.RESEARCH



O.RESEARCH will aim to develop human-like AI with superior performance, adaptability, and conceptual understanding, while prioritizing safety and ethical considerations through its Super AI Safety Committee. This multifaceted approach will encompass everything from core AI architectures to practical applications and services, driving toward a new frontier in artificial general intelligence.




O.RI

The heart of O.RI will leverage high-performance hardware infrastructure, including Cerebras' WSE-3 chip architecture, to ensure rapid inference and processing capabilities. This system will aim to access or maintain a registry of over 100,000 models, through HuggingFace's open-source model library, and will aim to connect to 200+ Inference-as-a-Service providers, such as OpenAI, Anthropic, and Cohere. O.RI will represent an industry-leading approach to AI query processing and model management, with a strong focus on ethical AI advancement, performance optimization, and enhanced model routing.


Key Features of O.RI

O.RI  Intelligent Routing⁠ A sophisticated AI routing system designed to optimize query processing and model selection.
O.RI BME  Benchmark Matching Engine⁠  For evaluation and optimized-routing
O.RI IA  Inference Awareness⁠  Optimized conceptualization of inferencing processes.
O.RI AF  Agentic Framework ⁠ For creating AI Agents or autonomous agentic systems.
O.RI AC  Atomic Chain of Thought⁠ Highly-specific reasoning and decision-making systems framework.
O.RI INFRA  Infrastructure and Security⁠ The hardware Infrastructure Hosting the ORI & models.
O.RI MLaaS  MLaaS API⁠ Machine-Learning-as-a-Service AI Query Processing System offering the O.RI as a service for enterprises.
O.RI API  Tools API and API Hu⁠ Tools APIs & TEE Cloud-functions Registry (HUB) powering the O.RI.


Why ORI is Needed

  • Single-model gap: Large language models vary in strengths; no single model excels at everything.
  • Existing routers rely on preferences: Many routing methods depend on human preference data, risking bias.
  • ORI advantage: Focuses on embedding-based segmentation, avoiding heavy human input.


How ORI Works

  • Vector embeddings: Transforms each query into a vector (Sentence Transformer), capturing semantic meaning.
  • Segmentation: Groups similar queries (K-Means/Agglomerative/KNN) to map them to dominant benchmarks.
  • Model selection: Identifies top-performing LLM for each cluster’s dominant task or benchmark.
  • Dynamic routing: Sends incoming queries to the most capable model automatically.


Key Outcomes & Takeaways

  • Performance gains: Up to +2.7 on MMLU, +1.8 on MuSR vs. best single models.
  • Broad effectiveness: Excels on diverse benchmarks like BBH, ARC, and MMLU.
  • Cost-speed balance: Maintains near-top token generation speed, keeps latency in check.
  • Scalable approach: Easily extends to new tasks and additional LLMs without re-engineering.