Projects

Work in progress

Beyond the four ready products, this is the other work worth showing. Each label is based on evidence: tests, CI and whether there is a live deployment.

Ongoing project: Under active development, with working code and tests; not a finished product.
Proof of concept: Demonstrates the idea end to end against fixtures, sample data or mock connectors; not deployed for customers.
Concept: Designed and started; early.

Open-source projects

The source for these is public on GitHub.

Proof of concept

BIM Manager agent

Takes an IFC model, finds the MEP clashes and clearance violations that are real, proposes moves that cite a clause, and verifies every proposal through three independent monitors (geometry, boundary, integrity). The model of record is never written to, and no language model is allowed near a clash verdict — enforced by a test.

Evidence: CI green on main (7 Sep 2026); 155 tests; 12 of 12 mutation probes killed; its own acceptance harness measures 9 of 30 checks passing, so it is not finished. No live deployment.

Source on GitHub (opens in a new tab)

Proof of concept

ThreadForge

An agent-native EPC piping digital thread: P&ID (DEXPI/Proteus-shaped XML) to topology graph, design volumes, piping artefacts (isometrics, PCF, quantities, test packs), AWP work packages and a 4D schedule, with every artefact regenerating when an upstream change cascades.

Evidence: v1.0.0 released 7 Sep 2026; acceptance harness 30 of 30; CI green on main. Runs against public DEXPI fixtures; its hard limits (for example proprietary CAD formats) are documented. No live deployment.

Source on GitHub (opens in a new tab)

Ongoing project

Cerebrum HotelOps

A sovereign hospitality operations platform: an operational reasoner and a guest-intelligence surface on one event bus, built on Cerebrum Blocks for UAE and generic markets, with fail-closed authentication. Version 2 is being developed in a private repository.

Evidence: version 2 has CI green on main (14 Sep 2026), with hotel-system connectors exercised against clearly labelled mocks. Version 1 is public; its CI on main is failing (12 Sep 2026). No live deployment.

Version 1 source on GitHub (opens in a new tab)

Private projects

The source for these is private, so there is nothing to link to. The descriptions and evidence come from each project’s own documentation and CI.

Proof of concept

FinanceOps

A multi-tenant finance operations and FP&A platform: planning, ingestion of actuals, reconciliation, governed approvals, AI-assisted retrieval over finance documents, and exports. Built on Cerebrum Blocks with FastAPI, React and Postgres with pgvector.

Evidence: CI green on main (23 Jul 2026) and on its latest change (5 Sep 2026). No live deployment.

Proof of concept

RetailOps

RetailOps — retail operations platform. Document-grounded answers with citations, audited retail actions (operations briefs, inventory risk, promotion compliance, reorder advice) and learning only from verified outcomes, with PDF, Word and Excel export.

Evidence: passed its own release certification as a pilot; point-of-sale, warehouse and EDI connectors run against functional mocks; CI green (11 Sep 2026). Not currently deployed.

Proof of concept

InsureOps

Insurance distribution and agency operations: producer records, agency hierarchy, channel routing, deterministic commission calculation, retention signals, incentive targeting, bordereaux analytics and compliance triage, built from the Store’s insurance kit.

Evidence: CI green on main (26 Jul 2026). Its own readiness review rules it not ready for production. No live deployment.

Proof of concept

Steward

A pilot runtime for private estate and remote asset operations, generated by the Factory: property registries, SOP-driven household operations, facilities and fleet, and a two-layer knowledge engine.

Evidence: CI green (5 Sep 2026). Its own readiness review allows a guarded pilot but rules it not ready for production; device (IoT) connectors are not built yet. No live deployment.

Proof of concept

TeamHub

A team workspace the Factory built from a one-sentence brief: multi-tenant roles, projects with documents, grounded questions and answers with citations, and a tamper-evident audit export.

Evidence: 31 of 31 end-to-end checks and 152 of 152 kernel tests pass in its test run; no CI configured and billing not connected. A Factory demonstration, not deployed.

Concept

The Level

An AI evaluation platform that turns domain expertise into versioned benchmark batteries, scored by independent graders, with a trust tier and an evidence pack for every result.

Evidence: scaffold stage (first acceptance milestone); CI green on main (3 Sep 2026). No deployment yet.

Want to collaborate on one of these, or see a demonstration? Get in touch.