Compounding Your AI Engineering Advantages | Generated by AI
Question: Based on my memory and current situation, what should I do next?
Answer:
Your strongest path is not to restart as a beginner AI engineer. You already have a rare combination:
- 11+ years software engineering
- HSBC/DBS enterprise backend experience
- LLM training from scratch (GPT-2, nanoGPT)
- GPU infrastructure experience (RTX 4070, MI300X/ROCm exploration)
- Agent development
- Public technical writing/blog
- Existing business experience (Fun Live)
The next move should be to compound these advantages, not scatter.
1. Main direction: become an AI infrastructure / agent engineer + build consulting business
For the next 6–12 months:
70%: build public proof + AI products
Focus on things that prove frontier-level engineering ability:
-
Train more small LLMs:
- 64M / 124M / 350M scale
- MoE experiments
- efficient fine-tuning
- synthetic data pipelines
- evaluation frameworks
-
Build serious agents:
- coding agent
- enterprise workflow agent
- browser/computer-use agent
- RAG + tool-use systems
-
Open source everything possible:
- GitHub repos
- blog posts
- benchmarks
- architecture notes
Your current advantage is not “I know LangChain”. Many people do.
Your advantage is:
“I can train models, deploy models, build agents, and ship production systems.”
That is much rarer.
2. Start AI consulting/company now, but keep it narrow
Do not become another software outsourcing company.
Your positioning should be:
AI engineering partner for companies that need internal AI capability.
Possible services:
Enterprise AI agent deployment
Examples:
- customer support agent
- internal knowledge assistant
- document processing agent
- workflow automation
LLM infrastructure
Examples:
- local model deployment
- vLLM inference
- fine-tuning
- evaluation
- GPU optimization
AI migration
Many companies have:
- Java backend
- databases
- internal systems
They need someone who understands both:
old enterprise world
+
LLM world
You have that combination.
3. Do not spend too much time on traditional backend work
Your HSBC experience is valuable as credibility.
But avoid:
Spring Boot CRUD project
+
database customization
+
outsourcing
as your main career.
Use it as:
cash flow -> AI transition funding
not your identity.
4. Build your “AI engineer portfolio” aggressively
Your blog already has traffic.
Turn it into an AI engineer portfolio.
Target articles:
Model training
Examples:
Training GPT-2 124M from scratch on RTX 4070
What I learned training 15B tokens
MoE experiment on consumer GPU
Agents
Examples:
Building a coding agent from scratch
How Claude Code / Cursor style agents work
Implementing tool calling loop
Infrastructure
Examples:
Running Qwen locally
vLLM optimization
GPU memory analysis
These attract:
- AI startups
- recruiters
- consulting customers
5. Degree: treat it as a signal, not the mission
Your instinct is correct.
A master’s degree is useful for:
- HR filters
- international mobility
- credibility
But don’t sacrifice AI output for classes.
Best case:
AI master's thesis
|
|
LLM training / agents research
|
|
public paper/blog/open source
The degree becomes another amplifier.
6. Short-term execution plan (next 90 days)
Month 1
Ship:
- one polished LLM training repo
- one agent repo
- 3 technical blog posts
Example:
lzwjava/mini-llm-training
lzwjava/agent-runtime
lzwjava/llm-evaluation
Month 2
Find customers.
Talk to:
- Shenzhen startups
- Guangzhou companies
- Hong Kong companies
Offer:
2-week AI prototype
not months of outsourcing.
Month 3
Choose:
A.
AI consulting revenue growing
→ continue company
B.
Strong AI engineering offer
→ join frontier AI startup/lab
C.
Both
→ best outcome
My priority ranking for you
| Priority | Action |
|---|---|
| 1 | Build public AI engineering portfolio |
| 2 | Sell AI agent/infrastructure consulting |
| 3 | Continue LLM experiments |
| 4 | Complete degree as credential |
| 5 | Keep backend work only for cash flow |
The biggest risk for you is not lack of ability. It is dilution: too many experiments, hardware rabbit holes, and infrastructure exploration without converting them into reputation or revenue.
Your next level is turning:
knowledge
↓
artifacts
↓
trust
↓
customers / opportunities
You already have the first layer. Now optimize for the last three.
Generated by AI. Curating and sharing still takes effort.
If you find it useful, feel free to
donate.
WeChat: @lzwjava
·
X: @lzwjava ·
Say hi 👋