A model that is not grounded produces plausible-looking output that does not survive design review. It suggests a beam section that ignores the governing load case, or a layout that violates a clearance rule the drawing set already encodes. The value is not in the model on its own — it is in how tightly the model is bound to real engineering inputs and real engineering constraints. Neilsoft builds AI applications across three related areas, each anchored to the systems and standards engineers work with every day.
Generative AI in Design
Two capabilities are routinely confused under this heading. They solve different problems, use different technology, and both fail without domain grounding.
Optimisation-driven
Generative Design
The approach in tools like Autodesk Fusion, where geometry is derived from goals and constraints — load cases, materials, allowable stress, preserved and obstacle geometry, and the intended manufacturing method (CNC, casting, or additive).
The solver returns candidate forms that satisfy the physics. Used well, it drives lightweighting, part consolidation, and manufacturability trade studies — but every candidate still needs an engineer to check it against the full requirement set.
Model-driven
Generative AI
Built on large language and diffusion models., It produces design options, specifications, technical documentation, and layout variants from prompts and existing project data.
On its own it is a fluent guesser. Made reliable, it is retrieval-grounded on your own standards, prior projects, and product data — so a drafted specification section or O&M document traces back to a source an engineer can verify.
Agentic AI and Custom AI Agents
A single answer is where most AI stops. An AI agent goes further: it plans a task, uses tools and data sources, and carries a multi-step workflow to completion. In engineering practice, that maps to the repetitive, checkable work that consumes billable hours:
Model and code checking
Validating a Revit or plant model against applicable standards, codes, and company templates, and reporting exceptions.
Data extraction and rollup
Compiling BOMs, attribute schedules, and quantities across drawings and databases into a single controlled output.
Submittal drafting
Assembling a first-pass submittal or transmittal from the model, the document store, and the project's requirements.
RFI triage
Reading incoming RFIs, classifying them, linking the relevant model elements and specs, and routing them to the right discipline.
MCP — Model Context Protocol
MCP (Model Context Protocol) is the open standard, introduced by Anthropic, for connecting AI models and agents to external tools and data in a consistent way. An MCP server exposes one of your systems to an agent through a standard interface — replacing brittle, one-off integrations with a governed, reusable connection.
The systems worth exposing are the ones that hold your engineering truth:
Neilsoft's capabilities
Generative and AI Assisted Design
Applying generative design and generative AI to real part, product, and layout problems.
Custom AI Agents
Agents that operate over engineering systems for QA, documentation, data extraction, and review workflows.
MCP Servers
Exposing PLM, Vault, CAD, BIM, and Forma data to AI agents through the Model Context Protocol.
Retrieval over Engineering Knowledge
RAG systems over standards, specifications, and project documents so answers are traceable to source.
Autodesk Platform Integration
Connecting AI workflows to design data through Autodesk Platform Services.
Applications in Engineering
Generative Design
Lightweighting, part consolidation, and layout optimisation.
Automated Drafting
Specifications, reports, and submittals generated from project data.
Model and Drawing QA
Agents that check against company and code standards.
Construction Workflows
RFI and submittal triage.
Knowledge Assistants
Grounded in a company's own standards and project history.
MCP Connected Agents
Live queries and updates against PLM and BIM data.
Watch us on YouTube
Join us on Linkedin