How We Use AI for Engineering Decisions

Project Dyson employs a unique approach to engineering analysis: querying multiple large language models and having them cross-review each other's analyses to build consensus on complex challenges.

Multi-LLM Consensus

Our analysis process involves three frontier AI models:

• Gemini 3 Pro: Google's advanced reasoning model
• GPT-5.2: OpenAI's latest generation
• Claude Opus 4.8: Anthropic's most capable model

Why Multiple Models?

Each AI model has different:

• Training data and knowledge bases
• Reasoning approaches
• Potential blind spots

By consulting multiple models and having them review each other's work, we can:

1. Identify consensus on well-established facts
2. Highlight areas of uncertainty
3. Discover novel approaches we might have missed
4. Cross-validate cost estimates and timelines

Our Process

1. Initial Analysis: Each model independently produces a detailed phase plan
2. Cross-Review: Each model reviews the other models' plans
3. Consensus Building: We synthesize areas of agreement
4. Gap Analysis: We highlight significant disagreements for human review

Limitations

It's important to note that AI analysis supplements, but doesn't replace, expert human judgment. All LLM opinions are reviewed by domain experts before being incorporated into our plans.

A Note on Model Versions Across Papers

The panel above reflects the models currently in use for BOM specs and deliberations. Several published papers were drafted and reviewed over a period spanning multiple model generations, so their reproducibility appendices name the specific models used at each stage rather than the current panel: most were reviewed by a February 2026 trio (Claude Opus 4.6, Gemini 3 Pro, GPT-5.2), with one paper's later revision rounds reviewed by a May 2026 trio (Claude Opus 4.7, Gemini 3.1 Pro, GPT-5.5 Pro). Each paper's own disclosure and appendix are the authoritative record for which models produced its results.