How ProjectVIEW AI Works on ERP Data

DANAOS ProjectVIEW AI works by retrieving live ERP data, structured by BoQ ↔ WBS ↔ Cost Codes, when a user asks a question or an agent runs. Then it passes that data to the model best suited to the task. ProjectVIEW AI is the AI add-on to ProjectVIEW ERP: agents and an assistant that work on ERP data. To do so, it combines domain-specific small language models with leading LLMs. However, every consequential action goes to a person for approval.

In short: deterministic ERP plus probabilistic AI. The ERP keeps the record. Then the AI reads it, explains it and suggests what to do next.

ProjectVIEW AI uses the data layer of ProjectVIEW ERP, the integrated ERP built on the BoQ ↔ WBS ↔ Cost Codes framework. There, every cost, commitment, quantity and progress record links to a BoQ item, a WBS package and a cost code. As a result, AI can compare like with like across estimates, purchase orders, subcontracts and actuals.

ProjectVIEW AI can also read data from outside the ERP:

• Third-party ERP such as SAP and Oracle.
• Schedules from Primavera P6 and MS Project, and BIM models (.ifc).
• EDMS, documents, email, contracts and IoT data.

ProjectVIEW AI also maps unstructured data to BoQ, WBS and cost codes. As a result, a contract clause or a machine fuel reading lands on the same structure as the budget. See native connectors and the BoQ software behind the structure.

ProjectVIEW AI retrieves the relevant ERP context at inference time, when the user asks the question or the agent runs. In other words, there is no retraining of models on client data. As a result, this has three practical effects:

• Answers use today’s figures, not the figures at the date of the last training run.
• ProjectVIEW AI applies role permissions at retrieval, so the model only receives what the user may see.
• Client data stays client data, so it does not become part of a shared model.

1. First, the user asks in plain language, by text or voice, in the ERP or through the MCP connector.
2. ProjectVIEW AI identifies the user and the role, and checks permissions.
3. Then it retrieves the permitted ERP records and documents for the question, linked by BoQ, WBS and cost codes.
4. After that, the model chosen for the task analyses the records and drafts the answer.
5. Finally, ProjectVIEW AI returns the answer with the figures and records behind it. As a result, the user can check it and drill into the ERP.

ProjectVIEW AI is model-flexible. Therefore, it chooses the model per use case.

Details are on the ProjectVIEW AI models page.

Every agent follows the same cycle:

1. Detect. The agent watches its process and spots a deviation, for example a quotation above the budget rate.
2. Investigate. It pulls the related records: the budget line, past rates, the supplier history, the WBS package.
3. Explain. It states in plain language what is off and why it matters.
4. Recommend. It proposes an action, such as negotiating, re-tendering or reallocating budget.
5. Request approval. It sends the recommendation to the person whose role owns the decision.
6. Execute. After approval, the ERP carries out the action.
7. Audit. The system logs every step, input and decision.

For a list of the eight purpose-specific agents, see the ProjectVIEW AI Agents page.

AI should operate inside enterprise governance, not around it. Therefore, each request passes through one chain: Identity → Role → Permission → Data → AI → Action.

• Identity: who is asking.
• Role: what the person is responsible for.
• Permission: which projects, records and approvals the role allows.
• Data: the system retrieves only the permitted records.
• AI: the model works on that data only.
• Action: the system executes anything consequential only after human approval.

ProjectVIEW AI does not send data to public consumer AI services. On Microsoft Azure, prompts do not go into model training. In addition, data residency options, encryption at rest and customer-managed keys are available. See security and governance.

Human approval sits between Recommend and Execute. In other words, ProjectVIEW AI can detect, investigate, explain and recommend without waiting. However, it cannot commit cost, change a contract value, certify progress or post a financial entry on its own. Instead, those decisions stay with the people who hold the authority in the ERP approval workflow.

The voice feature aims to let site teams work headless, without opening ERP screens. However, DANAOS confirms availability for your deployment during scoping. The pipeline design has six steps:

1. Voice capture on site.
2. Transcription to text.
3. Semantic interpretation of what was said.
4. Association to BoQ items, WBS packages and cost codes.
5. ERP fields populated, subject to validation and authorisation.
6. Reconciliation with existing records.

As a result, the pipeline aims to produce a speech-to-text daily site log for the user to review. However, voice recordings alone do not establish contractual entitlement or certified progress. People speak. AI interprets. ProjectVIEW ERP structures, validates and controls.