Ensure compliance of 3D designs: Complee (Since AI Hackathon)

Arthur Lagerlöf — Tue, 10 Nov 2026 15:27:44 +0000

This is a submission for the MLH x DEV Writing Challenge

Demo

Complee Live Demo

Try the ship design compliance platform. Click "Login" to open the demo and navigate to the 3D view to test the functions.

Open Complee

KEEP Team Page

Meet the team behind the project.

Meet the Team

At Since AI Hackathon we were proud to win both the Meyer Turku Challenge and Best Use of Vultr.

Summary of project

We built Complee, an AI-assisted ship design compliance platform that helps shipbuilding engineers identify potential regulatory gaps early by connecting 3D vessel models and engineering data with relevant regulations. Its modular architecture combines AI-powered information extraction, semantic regulatory retrieval and engineering assessment, with traceable findings and transparent handling of regulatory uncertainty. Built to support engineers rather than replace their judgement, Complee aims to reduce costly late-stage design changes. At Since AI Hackathon, we won the Meyer Turku Company Challenge and Best Use of Vultr.

Problem

Ship design involves thousands of interconnected components, systems and engineering decisions, each potentially subject to strict regulatory requirements. Discovering a compliance issue late in the design or production process can result in costly redesigns, material changes, delays and extensive coordination work.

Complee

Complee is a ship design compliance platform that connects vessel design data with regulatory requirements using AI to help engineers identify potential compliance gaps earlier in the design process.

The platform brings together three key capabilities:

Understanding ship designs

Complee processes engineering information from 3D models, CAD drawings and documents. It extracts relevant components, properties and relationships to build a structured representation of the vessel.

Finding relevant regulations

Using semantic search on a PostgreSQL database, Complee identifies potentially applicable requirements and connects them to relevant design information. Regulation records retain references to their supporting regulatory sources, making them easier to trace and review.

Assessing compliance

Complee compares available design information against identified requirements using Gemini API and deterministic models, highlighting potential compliance issues, missing information and cases requiring further engineering review.

Making regulatory uncertainty visible

Regulations are not always straightforward. Some requirements are objective and clearly defined, while others depend on context or require engineering interpretation.

Complee addresses this by having the ability to report that more information is needed to determine whether certain rules are followed. These are separate measures: an ambiguous requirement does not automatically mean a design is non-compliant. Making uncertainty visible helps engineers understand not only what might be wrong, but also where further interpretation or verification is needed.

Built for external integration

Complee is built around a modular architecture, allowing capabilities such as document parsing, 3D model interpretation, regulatory retrieval, AI providers and engineering calculation tools to evolve independently.

This architecture also creates opportunities for future extensions, including API-based integration with existing CAD and engineering software. The goal is to complement established engineering workflows rather than require companies to replace their current tools.

Complee is designed as an engineering decision-support system, not an autonomous certification tool. Engineers remain responsible for reviewing findings, interpreting requirements and verifying compliance.

Ultimately, we want to help ship designers identify potential problems earlier, when they are easier and less expensive to address.

Partner Technologies

Our main technologies include:

Python and FastAPI

Power the backend API and application logic, connecting the platform's different capabilities.

PostgreSQL and pgvector (In Tiger Data)

Store regulatory provisions and their vector representations, supporting semantic retrieval.

BGE-M3

Provides multilingual text embeddings for finding relevant regulatory provisions.

AI provider integrations

A provider-independent gateway allows different AI models and providers to be configured for specific tasks instead of tying the application to a single provider. Currently connected to Gemini 3.8 Flash.

3D model and document data processing

Extracts structured engineering information from formats including GLB/GLTF and DXF. Development of this workflow is still in progress, DOCX is the most reliable file type for input right now.

Vultr Object Storage

Stores large reference-model assets, which are served to the application when needed through a Node server on Vultr cloud compute.

One of many technical challenges: connecting regulations to engineering data

One of our biggest technical challenges was connecting two very different types of information.

Regulations are primarily written in natural language, whereas ship designs contain structured components, properties, relationships and geometric information. Simply asking an AI model whether a design is compliant would not provide the reliability or traceability required for an engineering application.

We therefore separated the workflow into information extraction, regulatory retrieval and compliance assessment.

AI helps interpret documents and engineering information, while structured data and deterministic engineering calculations can be used where appropriate. Findings retain supporting evidence and regulatory references so engineers can understand how potential issues were identified.

This separation also makes the platform easier to extend as new data sources, models and assessment capabilities are introduced, because all data about the design is stored in an identical structured format.

Vultr tech

We used Vultr cloud compute and object storage to stream the real ship model with optimizations (occlusion and size culling) of the R/V Taani from Oregon State University, with ~330,000 objects from the cloud to the browser. Our approach of streaming the model efficiently led to us winning a prize for the best use of Vultr at the hackathon.

Hackathon experience

Since AI Hackathon gave us the opportunity to tackle a challenging problem in a highly specialized industry: ship design compliance.

What made this project particularly interesting was bringing together AI, engineering data, 3D models and complex regulatory requirements. Instead of building another general-purpose AI assistant, we wanted to explore how AI could support a real engineering workflow without trying to replace what already works well.

Mainly we noted that applying AI to a domain with strict accuracy requirements meant applying generative AI only where it is strictly necessary and using deterministic methods when applicable.

We also learned the importance of handling uncertainty honestly. In engineering, a confident-looking answer is not necessarily a reliable one. A useful tool should communicate what it knows, where its information comes from and when an engineer needs to make the final judgement.

We are excited about the potential of Complee and look forward to continuing its development beyond the hackathon.

Feel free to comment if you have any questions!

Thanks to everyone who made Since AI Hackathon possible!

DEV Community: Arthur Lagerlöf