AI Trust Note:
Thinkproject AI in CDE NextGen

What is it?

Thinkproject AI provides grounded semantic search across PDF, image and Microsoft Office files in a CDE project, Users can ask natural language questions and receive:

  • AI-generated summary responses
  • References to supporting source documents
  • Context-aware conversational interactions within a session

Thinkproject AI also supports conversation and session management features, including:

  • Prompt history
  • Session history
  • Session pinning and unpinning
  • Session archive and unarchive
  • Session deletion
  • Rollback to previous prompts within a session

These capabilities are intended to improve continuity, traceability and usability of AI-assisted workflows within the CDE environment. They help teams locate, understand and act on project information more efficiently.

Thinkproject AI is intended to assist users in locating and understanding project information; it does not replace professional review, project governance, or document controls.

What data does it use?

Project-specific files stored in the CDE (e.g. drawings, specifications, schedules, reports, datasheets). Personal data may be present depending on the content in those files (e.g. names/emails in title blocks or comments). The quality and structure of project data directly influence retrieval quality and response relevance.

How is access controlled?

Thinkproject AI is designed to respect existing CDE document permissions through integration with the platform permissions API.

Users should only receive:

  • Search results
  • AI-generated summaries
  • Source references for documents they are authorised to access within the CDE.

Conversation and session management actions (such as delete, archive, pin/unpin, and rollback) apply only to the authenticated user’s accessible sessions and history.

Where it runs.

Thinkproject AI services are hosted on Google Cloud Platform and integrated into the NextGen CDE platform hosted on Microsoft Azure.

Depending on the contracted customer region, services and associated data may be hosted within:

  • United Kingdom
  • Europe
  • Australia

Technology.

Thinkproject AI currently uses:

  • Google Vertex AI embedding models
  • Google Spanner Graph DB
  • Google Datastore (vector search and indexing)
  • BigQuery
  • Gemini large language models (LLMs)

These services are used to:

  • Generate embeddings for semantic retrieval
  • Retrieve relevant document content
  • Generate grounded AI responses based on retrieved sources
  • Support conversational context and session continuity

Human Oversight.

Thinkproject AI is an assistive AI capability.

It does not automate project decisions or execute actions on behalf of users. Users remain responsible for validating outputs and reviewing referenced source documents.

Source links and citations are provided to support verification of generated responses. The use of AI-generated functionality is transparent within the user interface.

Quality + Monitoring.

We follow a QA approach prior to release and recommend regular regression testing, including when underlying model providers update base models.

Language Support.

Thinkproject AI currently supports semantic search and AI-generated responses in selected languages.

Current support status:

  • English — Supported
  • German — Supported
  • French — Pending datastore configuration
  • Spanish — Pending datastore configuration

For optimal results:

  • The user query, indexed document content, and generated response should all be in the same language.
  • Cross-language retrieval and translation-based querying are not currently supported.

Language support availability may vary depending on datastore configuration and deployment region.

⚠️ Limitations

Thinkproject AI may occasionally:

  • Return incomplete or partially relevant results
  • Generate inaccurate or overly broad summaries
  • Omit relevant documents from retrieval results

Current limitations include:

  • Support is currently limited to PDF, image, and Microsoft Office file formats
  • Responses depend on the quality and structure of indexed source documents
  • AI-generated answers should not be treated as authoritative without review of the underlying source material
  • Queries, indexed documents, and responses are expected to be in the same language for reliable retrieval quality
  • Cross-language semantic search is not currently supported
  • Retrieval quality may vary for scanned or low-quality documents
  • Very recent document updates may not appear immediately depending on indexing status
  • Rollback restores conversational context to a previous point in the session, but does not alter underlying project documents or external systems
  • Session and prompt history retention are subject to platform retention and lifecycle management policies