This is my first AI coding project - digital-twin. I used cursor implemented with GPT-5.3 Codex and reviewed by Anthropic Opus 4.6 and of course I reviewed it to my satisfaction. It has sleek look and the AI chat Q&A quality (stick to source of truth) is pretty good. I think that I gradually gain some confidence on AI coding and consider AI as my assistant and extends my capability. However, I and AI coding might be in Honeymoon period :. More to come. Let's see.
The followings are two screenshots.
Both 'tutorial.md' and 'review.md' are under /web sub-folder.
When I have a chance, I will review Ed's Production Track course material to see if I can put together Vercel configuration and deploy to Vercel.
This is the history of my instructions for my own record too:
Please build me a professional website running locally. My linkedin profile is profile.pdf Make my website stunning: enterprise meets edgy. It should include about me, my career journey and a link to my portfolio (for future). Iterate to make it as slick and professional as possible, Use Next.js.
That's great. Please now add the ability to have an AI chat with a "digital twin" that can answer questions about my career. Please use OpenRouter. My OPENROUTER_API_KEY is in ".env" file. Please use a model name "openai/gpt-oss-120b". Go ahead to plan and make changes. Make sure it works. Let me know when it is ready.
Look great!! I see the problem. Yes, Please render markdown in response and also add chat streaming.
K. Can you use read the full content of profile.pdf as the DIGITAL_TWIN_KNOWLEDGE instead? The current DIGITAL_TWIN_KNOWLEDGE is incomplete.
Yes, Your response now is derived from the full content of "profile.pdf". One more issue. I am seeing HTML "<br>" in the response. Can you render HTML '<br>' as a single line break in the response? Thanks.
1. As you can see, it does not format Agents section in 3 column format well. You can simplify not to use a format of 3 columns as long as the response is readable and displays cleanly.
2. You need to give LLM much explicit instructions: Stick to the topic. For example, do not retrieve and include the content from my "AI Engineer Production Track: Deploy LLMs & Agents at Scale" course if I ask for the details of "Agentic RAG" project only.
Please write a comprehensive tutorial in markdown called tutorial.md that is suitable for a beginner in frontend coding. Include a summary of technology, a high-level walk through, and detailed code review with code samples. End with 5 suggestions for ways that codes could be improved based upon a self review.
Please do a comprehensive code review of this project and write the result to review.md including any remedial actions needed. DO NOT actually change any code.
There are 3 high-priority remedial action plan that I like you plan and make changes.
1. Add rate limiting to POST /api/digital-twin. A simple in-memory sliding window (e.g., max 20 requests per minute per IP) prevents runaway credit burn.
2. Add a test framework and initial tests. Install vitest. Write unit tests for extractSection, buildFocusedProfileContext, extractProjectEntries, pickBestProjectEntry, and normalizeAssistantMarkdown.
3. Add AbortSignal.timeout(30_000) to the OpenRouter fetch to prevent hanging requests.
Yes, can you add a short TESTING.md, with instructions and examples for running a specific test file or test case.
