The Verde Archive bot

Lately I’ve been writing a lot on LinkedIn about Product, AI, and related topics. But LinkedIn is terrible at retrieving old posts. Finding something I wrote months ago is nearly impossible unless I scroll endlessly.

The obvious solution is to move my content to a proper platform like a blog or CMS. But I wanted a pragmatic solution now, without investing days migrating and structuring everything.

So I decided to build a lightweight AI-based lookup tool. To be clear, this is not a production-grade retrieval system. It’s a simple productivity prototype to solve my immediate need. Here is what I did:

• Created an Excel file with all my posts titles and links.
• 🔒 No company docs or private data. Only my public post titles and links.
• Asked ChatGPT to generate short descriptions, keywords, and tags for each post (providing my original article, no LinkedIn scraping involved)
• Converted it into JSON (ChatGPT handles structured data better this way).
• Created a custom GPT that uses my posts’ JSON data as static structured context for prompt-based retrieval. This is not a full RAG pipeline with embeddings and vector search, but a pragmatic lookup approach within ChatGPT.
• Instructed the custom GPT on accepted questions and ‘behavior’ (If curious let me know and I will give details in another post).

The result? An AI assistant that retrieves any of my posts instantly. By topic, description, or tags. No more wasted time searching. Immediate recall, every time.

Is it working well? Just built it, still testing.

In early use, it retrieves posts by keyword instantly, saving me time and mental load.

Why does this matter?

Right now, this is just a personal productivity hack. But it points to a bigger shift:

• AI as an extension of your personal knowledge base
• Instant retrieval and summarization of your own insights
• Personal content curation without needing full CMS migration

I believe every knowledge worker will soon have AI tools tailored to their own content, workflows, and mental models. This was my first step in building mine (or at least experimenting).

Next steps

Automate it. Currently, I update the JSON manually whenever I post new content. It’s not much work, but it’s annoying and prone to delays. When I automate the sync, I’ll use the official API. If that isn’t possible, I’ll add it to my pre-post workflow.

Even if I move to a proper content platform later, this AI approach will remain useful as a fast retrieval layer.

Why stop at retrieval?

Some people feed all their content into a model and let a “digital twin” answer for them. I won’t. I enjoy the debate, the fresh viewpoints, the human back-and-forth. This assistant is only a memory aid that helps me surface my own work faster, it is not a stand-in for me. Others may choose full chatbot clones, but that isn’t for me. That’s why I set guardrails for the custom GPT.

Happy to hear your feedback or dive into details. Just drop a comment below!