From 37-Year Marine Engineer to AI Model Evaluation Expert: My Career Transformation Journey

Introduction: Why Transform at 59?

My name is Ma Weize. I'm 59 years old and a professor-level senior marine engineer. From my university graduation in 1988 to 2024, I've worked for 37 years in ship design, high-speed craft development, and propeller design. At this life stage, many would choose to enjoy the stability before retirement, but I made a decision that surprised everyone around me: a complete transformation into AI model evaluation and safety alignment.

This wasn't an impulse. Over the past year, I've intensively used large language models like Claude, Deepseek, and ChatGPT to assist my work, personally experiencing how AI changes engineering design workflows. More importantly, I discovered a unique opportunity: my 37 years of engineering experience is exactly what AI model evaluation needs most.

I. Discovering a New Direction: From User to Evaluator

1. Insights from a Heavy AI User

In early 2024, I started using Claude Code to help develop ShipTechAI—a ship propeller design tool based on the Rolla SPP database. Through this process, I gained deep comparative knowledge of different AI models:

• Deepseek: Excellent in engineering calculations and data processing, cost-effective, particularly good at implementing complex mathematical formulas
• Claude: High code quality, strong comprehension, but limited depth in certain specialized domains
• ChatGPT: Balanced general capabilities, but accuracy in professional engineering applications requires repeated verification

Through daily use, I began pondering: How reliable are these models in industrial scenarios? How much human verification do their outputs require? How can we systematically evaluate their performance?

2. The Unique Value of an Engineer's Perspective

After deeply exploring LLM model evaluation and Red Teaming, I found my background offers unique advantages:

1. Real Scenario Understanding: I know how engineers actually use AI in their work and which scenarios are most error-prone
2. Professional Knowledge Depth: Ability to design complex domain-specific test cases, not just simple generic questions
3. Systematic Thinking: 37 years of engineering project management experience makes me adept at establishing evaluation frameworks and methodologies
4. Risk Awareness: The engineering field's strict requirements for safety and reliability align perfectly with AI safety alignment goals

II. Transformation Path: How Engineering Thinking Accelerates AI Learning

1. Phase One: Learning Python from Scratch (10 Days)

Many think learning programming at 59 is impossible. But I proved with 10 days—from zero foundation to independently developing the ShipTechAI full-stack application—that engineers' learning capability is powerful.

"I didn't follow traditional tutorials to learn Python. I started directly with projects. When I didn't know something, I asked Claude, requiring it to explain the principles for each feature. This 'learn-by-doing' approach let me master content that typically takes months in just 10 days."

Key Methods:

• Project-driven learning: Don't study syntax, start writing ShipTechAI directly
• Claude as mentor: Require explanations for every line of code, understand principles rather than memorize
• Engineering mindset: Think of code as engineering design—modular and systematic
• Rapid iteration: Implement functionality first, optimize later, don't pursue perfectionism

2. Phase Two: Deep Dive into LLM Principles and Evaluation Methods (1 Month)

After mastering Python, I systematically studied LLM working principles, prompt engineering, and model evaluation methodologies. In this phase, my engineering background proved valuable again:

• Comparative Testing: Like testing different propeller designs, I established systematic model comparison testing frameworks
• Quantitative Evaluation: Engineers are used to speaking with data; I created quantitative metrics for different models' performance
• Boundary Condition Testing: The extreme case testing mindset from engineering design corresponds to AI Red Teaming
• Failure Mode Analysis: FMEA (Failure Mode and Effects Analysis) from ship design applies perfectly to AI safety assessment

3. Phase Three: Practical Project Accumulation (Ongoing)

I accumulated evaluation experience through real projects:

1. ShipTechAI Development: Deep comparison of Deepseek and Claude's performance in engineering applications
2. Professional Domain Testing: Designing a marine engineering-specific model evaluation case library
3. Industrial AI Safety Research: Studying LLM reliability in high-risk scenarios like industrial control and design calculations
4. Blog Sharing: Documenting the transformation process to help other traditional engineers understand AI

III. Key Insights: Engineers' Unique Advantages in AI Evaluation

1. Real-Scenario Evaluation Cases

Most AI evaluations are based on generic tasks (translation, Q&A, code generation). But industrial AI's real challenge lies in complex scenarios in professional domains. For example:

2. Systematic Evaluation Framework

Systems engineering methods from project management apply directly to model evaluation:

• Requirements Analysis: Clarify evaluation objectives (accuracy? safety? efficiency?)
• Test Planning: Design case sets covering various scenarios (normal, boundary, exceptional)
• Quantitative Metrics: Establish measurable evaluation standards (error rate, response time, consistency)
• Risk Assessment: Identify high-risk scenarios and test them intensively
• Continuous Improvement: Establish feedback mechanisms, continuously optimize evaluation methods

3. Industrial AI Safety Practical Experience

Marine engineering is a high-safety-requirement field, giving me natural sensitivity to AI safety:

• Severe Failure Consequences: Ship design errors can cause maritime disasters; AI errors in industrial control are equally fatal
• Multiple Verification Mechanisms: Engineering design's review processes correspond to multi-layer verification of AI outputs
• Conservative Safety Margins: Safety factor thinking in engineering design applies to fault-tolerant AI system design
• Documentation Traceability: Engineering projects' strict documentation requirements correspond to AI decision explainability needs

IV. Practical Results: The ShipTechAI Project

1. Project Overview

ShipTechAI is my first complete project after transitioning to AI. It's a ship propeller design tool based on the Rolla SPP database. This project not only validated my technical capabilities but also became an experimental platform for studying AI model performance in engineering applications.

2. Technical Implementation

• Data Processing: Digitized 101 design points from the Rolla database, established a 3D interpolation model
• Algorithm Implementation: Integrated B-Troost chart method and Rolla SPP database
• Intelligent Diagnostics: Automatically identifies potential issues and provides recommendations based on design parameters
• Full-Stack Development: HTML/CSS/JavaScript frontend + Python backend + mathematical models

3. Deep Application of AI-Assisted Development

During ShipTechAI development, I deeply compared different AI models' performance:

V. Future Direction: Industrial AI Evaluation and Safety Alignment

1. Short-term Goals (6 Months)

• Build a marine engineering-specific LLM evaluation case library (500+ test cases)
• Systematically compare Deepseek, Claude, Qwen and other domestic models in industrial scenarios
• Research prompt engineering best practices in professional domains
• Participate in or apply for AI safety assessment-related projects

2. Medium-term Goals (1-2 Years)

• Join AI model development teams as an industrial domain evaluation expert
• Establish industrial AI safety evaluation methodologies and standards
• Train other engineers to transition to AI evaluation
• Present research findings at AI safety conferences

3. Long-term Vision

Promote a new "Engineer + AI Evaluator" career development path. Traditional engineers possess rich professional knowledge and practical experience, which is exactly what AI model evaluation needs most. Through my transformation journey, I hope to prove:

• Age isn't a barrier to transformation; experience is actually an advantage
• Engineering thinking remains powerful in the AI era
• Traditional industry experts are key forces in AI safety
• Lifelong learning is the only constant in career development

VI. Advice for Traditional Engineers Considering Transformation

1. Mental Preparation

• Embrace Change: AI isn't a threat but a tool. Rather than worry about being replaced, learn to use it
• Trust Experience: Your professional knowledge is more valuable in the AI era; don't underestimate yourself
• Stay Curious: Like when you learned engineering, maintain curiosity about new technologies
• Dare to Try: Age is just a number; learning ability is the core competitiveness

2. Learning Path

1. Step One: Become a heavy AI user (3 months)
   • Use Claude/Deepseek daily to solve real work problems
   • Record which scenarios AI performs well in and which it doesn't
   • Learn prompt engineering to improve AI usage efficiency
2. Step Two: Learn Python basics (1 month)
   • Don't read tutorials; start projects directly with AI assistance
   • Choose automation tool development related to your job
   • Require AI to explain principles for each feature
3. Step Three: Deep dive into AI evaluation (ongoing)
   • Systematically learn LLM principles and evaluation methods
   • Design evaluation cases for your professional domain
   • Compare different models' performance and document results
   • Participate in open-source AI evaluation projects

3. Pitfalls to Avoid

• ❌ Starting with basic syntax when learning programming (too slow, easy to give up)
• ❌ Pursuing perfect code quality (implement functionality first, optimize later)
• ❌ Thinking you're too old to learn (I'm 59 and did it; you can too)
• ❌ Expecting to master all knowledge at once (learn by doing, accumulate gradually)
• ❌ Ignoring professional advantages (your domain knowledge is your greatest asset)

Conclusion: Lifelong Learning, Never Stop

My 37-year engineering career taught me one thing: technology changes, but problem-solving thinking methods remain constant. From ship design to AI model evaluation, they appear to be completely different fields, but the essence is the same: how to establish systematic methodologies, how to find optimal solutions under complex constraints, how to ensure system safety and reliability.

Transforming at 59 isn't an endpoint but a new starting point. The AI era offers traditional engineers unprecedented opportunities: our professional knowledge, systematic thinking, and engineering experience are exactly what AI model evaluation and safety alignment need most. Age isn't a barrier; lifelong learning is the only constant in career development.

"When I learned Python in 10 days, developed ShipTechAI in 1 month, and became a practitioner in AI evaluation in 3 months, I realized: learning ability doesn't decline with age, as long as you maintain curiosity and drive to act."

If you're also a traditional engineer thinking about how to face the AI era's challenges, I want to tell you: now is the best time. AI isn't here to replace you; it's a tool that makes your experience more valuable.

About the Author:
Ma Weize, professor-level senior marine engineer with 37 years of experience in ship design and high-speed craft development. Currently focused on LLM model evaluation, AI safety alignment, and industrial AI application research. Creator of the ShipTechAI tool.

📧 Contact: jason15994264083@gmail.com
🌐 Website: www.shiptechai.com
🚁 ShipTechAI Tool: Try Online

← Back to Blog