Projects & Architecture
A register of Manifold Grace software ideas, existing demonstrations and venture opportunities, with a baseline and an acceptance test for each main build.
AI model architecture families · 1,000 proposed configurations · Modular platform architecture · Assurance pathways · Academic crosswalk · Project register · 2027 opportunities · Learning outcomes · Related work
AI and machine-learning development — distinct types
One identifiable architecture family, algorithm or system method per project. The current four-level register distinguishes 17 active application and capability categories, 10 learning paradigms, 36 methods and frameworks, and 53 model or system-architecture records. Nine historical entries remain linked to their replacements. A model-architecture family, a system-integration pattern and a specific model topology are explicitly labelled as different kinds of Level 4 records. The current v1.6 register holds 116 active entries, adding nine historically documented search, symbolic planning, optimisation and graph-modelling approaches. SRC-009 provenance and v1.4 scope descriptions have also been reconciled. Coverage remains incomplete and project implementations have not been independently verified.
Browse corrected four-level AI taxonomy · Earlier three-table classification · Architecture types · Methods and learning paradigms · Linked projects and paper initiatives · Matching Excel-compatible table (v1.6) · Original scenario research
Architecture, legal and security assurance learning pathways
Linked project specifications for studying original papers and permitted code sources, reconstructing architecture for teaching, testing compliance and cybersecurity safeguards, and choosing appropriate academic or professional paper formats. Fifteen candidate project families and 18 publication formats are distinguished, with beginner, intermediate and advanced pathways; proposals are not described as completed implementations.
Explore architecture, assurance and research project pathways · Paper and publication options · Difficulty and repository organisation
Live archival reference implementation and three research papers
A reusable source-linked publication archive generates daily, weekly, monthly, annual and complete PDF catalogues, alongside accessible browser navigation and JSON/CSV records. The implementation indexes dated public site records rather than copying external articles. Three separately labelled working papers address design-science architecture, legal-policy options and the use of living evidence archives in higher education and advanced AI evaluation.
Explore daily-to-annual PDF archive · Research papers and source-code lineage · Reusable generator
Academic crosswalk
Read each row from left to right: project, syllabus and reading, programme outcome, then the specific application a reviewer can inspect. Project links lead to the existing source pages. The exercises describe evidence to produce or review.
Sources and interpretation
Weekly topics and reading assignments come from the History and Foundations knowledge map and selected reading audit. Cross-week applications borrow a reading from another point in the syllabus. Outcome numbers follow the official 2024/25 programme specification. The links are proposed alignments for academic review. LO8 needs independent research evidence and LO9 needs documented collaboration.
Evidence gap projects
Public Gap Register · Public procurement evidence gaps · Institutional evidence case study · Clinical trial traceability project
Project register
Start with independent chat, automation, CRM, ERP and compliance. Each exercises a different architecture. The linked source pages document the work so far. Starter and test columns describe proposed preparation.
| Project | Manifold Grace source | Starter or architecture | First acceptance test | MA outcomes |
|---|---|---|---|---|
| Independent ChatGPT-style interface | Source page | WebLLM minimal chat example | Run one multi-turn chat, export it, and test cancellation and unsupported devices | 1, 7 |
| Business process automation | Source page | Node-RED flow with synthetic events | Route an enquiry through approval and replay a duplicate event | 1, 3, 4 |
| CRM and Salesforce-style pipeline | Source page | Frappe CRM sandbox | Move a synthetic lead from enquiry to opportunity with role checks | 1, 2, 3 |
| ERP and resource planning | Source page | ERPNext sandbox | Trace a test order through stock, invoice and reconciliation | 1, 3 |
| Compliance and continuous AI assurance | Source page | Django and SQLite control register | Link a control to evidence, a failed test, remediation and sign-off | 2, 4, 5 |
| Evidence verification and contradiction API | Source page | FastAPI and versioned evidence fixtures | Benchmark claim coverage, contradiction and justified abstention | 1, 3, 8 |
| Research discovery and literature-gap audit | Source page | Python search pipeline and SQLite | Compare retrieval against a fixed literature set and test one proposed gap | 6, 8 |
| Cross-conversation context and private memory | Source page | Local database and retrieval adapter | Find a consented record, apply a correction and verify deletion | 1, 4, 5 |
| Historical as-of retrieval and correction propagation | Source page | Versioned relational records and API | Reconstruct a claim at two dates and propagate a superseding record | 3, 5, 8 |
| Document-drop classifier and evidence organiser | Source page | Python ingestion, local metadata and file preview | Classify mock files and preview filenames before applying any change | 1, 3 |
| Contributor provenance and innovation attribution | Source page | Relational contribution register | Record origin, revisions and an attribution dispute | 4, 5, 9 |
| Agency.ai agent workspace | Source page | Workflow engine and explicit approval queue | Run a simulated service request with a reviewable action ledger | 1, 4, 5 |
| Public-interest research and societal resilience adviser | Source page | Retrieval and scenario dashboard | Compare two policy scenarios and cite assumptions and evidence gaps | 2, 6, 7, 8 |
| UK public bodies and oversight directory | Source page | Static directory plus versioned JSON graph | Trace a body's remit and oversight across a dated reorganisation | 5, 7, 9 |
| Learning outcome, viva and competence verification | Source page | Rubric and interactive quiz app | Compare rubric results on matched sample responses and collect reviewer agreement | 2, 3, 8, 9 |
| Prior learning and skills evidence portfolio | Source page | Evidence-to-outcome matrix and review queue | Map one sample portfolio to criteria with a human review decision | 4, 7, 9 |
| Media scheduling and publishing planner | Source page | Calendar and rights metadata | Draft a schedule from licensed sample assets and flag a rights conflict | 2, 4, 7 |
| Message reply and personal assistant | Source page | Draft queue and mock connectors | Generate and export a reply after human review | 1, 4 |
| Multimodal perception and decision support | Source page | Image inference and review dashboard | Compare results under changed lighting with uncertainty and human override | 1, 3, 8 |
| Architecture comparison and graph mapping | Source page | Static diagrams and dependency graph | Compare three system designs against the same requirements | 1, 6, 7 |
| Model evaluation, bias and reliability | Source page | Python benchmark runner and reporting UI | Measure disagreement, subgroup errors, latency and cost on synthetic cases | 2, 3, 8 |
| Opportunity window and expenditure calculator | Source page | Browser calculator with test scenarios | Show how changing assumptions alters rankings and costs | 1, 6, 7 |
| Ga community development metrics | Source page | Static dashboard and aggregate dataset | Trace one aggregate measure to a dated source | 2, 7, 9 |
| Family history evidence explorer | Source page | Versioned graph and privacy-filtered views | Compare two hypothetical lineage claims with source strength | 4, 7, 8 |
| Distinct Histories introduction app | Source page | Static interactive mock-up | Test consent, disclosure and accessible onboarding with fictional profiles | 2, 4, 7 |
| Intentional dating introduction brief | Source page | Local form and export | Generate a fictional brief and test data minimisation | 2, 4, 7 |
| Creative portfolio and interactive globe | Source page | Static media portfolio and WebGL prototype | Test keyboard navigation, reduced motion and mobile performance | 3, 7 |
| Botanical collection and product catalogue | Source page | Static catalogue and simulated basket | Compare accessible product discovery and mock checkout | 3, 7 |
| Research manuscript template library | Source page | Static download catalogue | Download and open a versioned sample template with provenance | 7, 8 |
| Job application and career support | Source page | Local CRUD tracker and document export | Record and export a fictional application and its follow-up | 3, 7 |
| Equitable living and public initiatives directory | Source page | Static taxonomy and source register | Find an initiative by geography and verify its source | 2, 5, 7 |
| Financial reconciliation and payments sandbox | Source page | Mock webhook listener and ledger | Reconcile duplicate and late payment events using synthetic transactions | 1, 3, 4 |
| Legacy connector and data migration lab | Source page | CSV fixtures and a Python import adapter | Re-run an import without duplicating records and produce an exception report | 1, 3 |
| Technical demonstration portal | Source page | Static catalogue with runnable demos | Check source, limitations and test instructions for each demo | 7, 9 |
2027 venture opportunities
All 28 opportunities from the venture portfolio. These are candidate offerings. Several can use the same project baseline.
| Opportunity | Potential offering |
|---|---|
| 01. AI evidence verification API | B2B API for claim-to-source coverage, contradictions and dated provenance |
| 02. AI system architecture mapping | Diagram-to-dependency maps with implementation discrepancy reviews |
| 03. Agent identity and permission audits | Inventory agents, credentials and authorisation boundaries |
| 04. Coding-agent quality assurance | Tests, regression monitoring and change provenance for generated code |
| 05. SME workflow implementation studio | Industry-specific automation with training, rollback and support |
| 06. AI procurement due-diligence reports | Evidence-based supplier comparison, data handling and exit-risk dossiers |
| 07. AI context and knowledge governance | Versioned organisational knowledge, retrieval access and correction histories |
| 08. Model and agent expenditure observability | Dashboards for model costs, latency and workflow value |
| 09. AI assurance documentation service | Human-reviewed risk registers, evaluation evidence and audit packets |
| 10. AI literacy and verification training | Applied training in checking outputs and safe workplace integration |
| 11. Accessible AI interface auditing | Test accessibility and usability of AI-supported public or business services |
| 12. Human oversight workflow tools | Approval queues, escalation, sign-off and traceability for AI actions |
| 13. Secure AI adoption for charities | Low-cost setup, staff guidance and data-safe workflows |
| 14. AI cybersecurity evidence triage | Defensive summarisation and prioritisation of authorised alert data |
| 15. Provenance and authenticity workflow | Content origin, revisions, signatures and source verification |
| 16. Research literature and gap audit assistant | Falsification-led evidence maps, overlap detection and reproducible reviews |
| 17. AI-enabled public-systems directory | Versioned public-body relationships, functions and oversight routes |
| 18. AI learning-outcome verification | Assess student understanding rather than AI-generated assignment polish |
| 19. Physical AI simulation evaluation | Testing robotics and simulated workflows without selling generic coding toys |
| 20. 2027 opportunity window intelligence | Monitor adoption, unmet need, competitor substitution and window-closing signals |
| 21. Synthetic data & edge-case generation | Validated rare-scenario datasets for model and robotics evaluation |
| 22. Legacy-to-AI bridging & middleware | Permissioned data connectors and translation layers for older enterprise systems |
| 23. AI liability and risk analytics | Incident evidence, control assessment and risk telemetry for insurers |
| 24. Human-in-the-loop escalation routing | Uncertainty-triggered human handoffs, contextual review and approval records |
| 25. On-premise & sovereign AI deployment | Local or residency-controlled AI stacks with documented security boundaries |
| 26. AI-native application consolidation | Subscription inventory, workflow consolidation and value-based software rationalisation |
| 27. Machine-readable brand & commerce optimisation | Structured product data and verified metadata compatible with agentic shopping workflows |
| 28. Post-automation workforce reskilling | Evidence-based scenarios for human review, system judgement and exception handling |
Learning and preparation
The 2024/25 programme specification lists nine outcomes. The numbers above refer to AI foundations (1), ethical evaluation (2), case studies (3), applied ethical reasoning (4), policy and regulation (5), critical discussion (6), communication (7), independent research (8), and interdisciplinary collaboration (9). Outcome 9 needs actual participation and documented feedback.
For each build, record the upstream source and version, licence and notices, installation steps, synthetic fixtures, a working baseline commit, and an experiment branch. Add a data-flow diagram, an acceptance test, a note on what broke, and a record of what you changed. Capture the test result before linking a working demo.
Use WebLLM for browser inference and GitHub Pages for static demos. Run Node-RED, Frappe CRM, ERPNext and Django locally for experiments that need a backend. Verify the specific release and licence before distributing a modified baseline.
Test expected challenges early: replacing a model or provider, recovering from failed workflows, retaining evidence versions, enforcing access rules, exporting data, and keeping the interface accessible.