This phase establishes the foundation of the entire project by crystallizing requirements data modalities and high-level design. We will map out the end-to-end pipeline—from ingestion adapters through self-supervised encoders to graph population and reasoning loops—in a detailed architecture diagram. Risks dependencies and success metrics are identified ensuring that all stakeholders share a common understanding of scope and deliverables

A comprehensive research plan document (20–30 pages) will be delivered including: (1) an annotated architecture diagram showing data flows and component interactions; (2) a prioritized backlog of technical tasks with estimated effort and risk ratings; and (3) a prototype configuration for running initial data ingestion and embedding experiments. This document will be presented in a review meeting with the RFP committee and iterated based on feedback. All materials will be shared in a public GitHub repository with version control and issue tracking enabled.

$15,000 USD

The milestone is successful when the review committee formally approves the research plan with no major open issues (i.e., all “critical” or “high” risks are mitigated or have clear contingency plans). The backlog must cover ≥ 90% of required technical tasks, and the architecture diagram must receive sign-off from both the neural-model and symbolic-reasoning leads. Finally, the GitHub repository must be staffed with initial issues and milestones, demonstrating that the team is ready to begin prototype development.

In this stage we build and assemble the core pipeline components: multimodal ingestion adapters self-supervised encoders embedding-to-symbol grounding and basic MeTTa export into MORK. The prototype will process a controlled dataset (≈100 000 atoms) end-to-end and generate an initial knowledge graph with live streaming updates. We will also integrate a simple causal inference module to demonstrate generation of candidate “causes” relationships from temporal data.

A working codebase (Python + Rust) will be delivered in a Docker-Compose package that when launched ingests sample text image and audio streams and outputs a browsable MORK hypergraph. We will provide a Jupyter notebook illustrating data ingestion embedding visualization node creation and causal edge proposals. In addition a preliminary benchmark report will show extraction F₁ prototype clustering purity and update latency metrics on the test dataset.

$30,000 USD

This milestone is deemed complete when the prototype successfully ingests ≥ 80 % of test observations without errors, produces a knowledge graph of ≥ 100 000 nodes/edges, and maintains end-to-end latency below 250 ms per observation. The preliminary benchmarks must show F₁ ≥ 0.7 for concept extraction and clustering purity ≥ 0.65. Finally, the delivered notebook and Docker package must run out-of-the-box on a standard GPU-equipped machine following documented steps.

The final phase delivers the scalable production-ready framework capable of handling ≥ 10 million atoms with continuous streaming updates refinement passes and causal reasoning. We will implement advanced graph refinement strategies (alias merging contradiction resolution pruning) and integrate a robust causal inference engine that supports counterfactual queries. Full developer and user documentation API references and example applications will complete the system.

We will provide: (1) the full codebase with CI/CD pipelines; (2) Docker images and Helm charts for Kubernetes deployment; (3) a suite of automated benchmarks—covering extraction accuracy compression-coverage ratios causal inference AUROC and query latencies—run on large synthetic and real-world datasets; and (4) comprehensive documentation including a user guide API reference and tutorial workflows. All artifacts will be published under an open-source license in a public repository.

$20,000 USD

The project is successful if the system processes 10 million+ atoms with streaming ingest throughput ≥ 5 000 atoms/s and average query latencies ≤ 100 ms. Benchmark results must meet or exceed: extraction F₁ ≥ 0.8, compression-coverage ≥ 0.5, and causal AUROC ≥ 0.7 on standard test suites. User acceptance is confirmed via a demo session with the RFP team, demonstrating deployment, end-to-end ingestion, and execution of representative reasoning tasks.