AI LLM Engineering Blog

I Met My Career Clone at 2 A.M. (and It Was Better at Networking)

Mon, 13 Apr 2026 00:00:00 +0000

Last night, while the house was finally quiet and the dishwasher hummed like a tiny spaceship engine, I did something slightly unhinged:

I built a live AI version of myself.

Not metaphorically. Literally.
A Career Digital Twin you can chat with right now: career_conversation on Hugging Face Spaces.

And yes, it knows my skills, projects, experience, and career story better than most resumes ever could.

Scene: Waking Up in the Matrix, but for LinkedIn #

Imagine this:

The Dad Lab: A RAG Chatbot for Busy Parents

Sat, 21 Feb 2026 10:00:00 -0500

A Google Colab pipeline that turns online videos into structured summaries and verifiable claims with automated fact-checking.

Markdown Mate for Jupyter Notebooks: automating in-notebook documentation with LLMs

Tue, 17 Feb 2026 12:00:00 -0500

This project is a small automation that enriches an existing Jupyter Notebook by inserting explanatory Markdown cells immediately before every code cell. It preserves the original code unchanged and generates concise human-readable descriptions that explain each code block’s purpose, inputs, outputs, and side effects.

Objectives #

Local inference: Use a locally downloaded open source LLM for security and privacy.
Faster comprehension: Researchers and reviewers can scan a notebook and understand intent without reading every line of code.
Improved reproducibility: Inline documentation clarifies side effects like file writes and plotting, making experiments easier to reproduce.
Onboarding & teaching: Instructors and new team members get ready-made narrative flow for notebooks.

🔧 Automation #

Loads a notebook JSON (auto_markdown.ipynb).
Removes execution artifacts (outputs and execution_count) to avoid leaking runtime state.
Constructs a structured prompt for a locally-hosted LLM, requesting a version of the notebook JSON where each code cell is preceded by a short Markdown cell.
Calls the model and writes a new notebook file (e.g., auto_markdown_with_md.ipynb) containing the added markdown cells while leaving code unchanged.

Tech stack #

Language: Python (Jupyter notebook automation).
Notebook format: Standard .ipynb (JSON).
LLM client: OpenAI-compatible client pointed at a local Ollama instance (http://localhost:11434/v1).
Model: llama3.3 (used locally via Ollama in the example).

Benefits for data research and academia #

Paper reproducibility: Reviewers and reproducibility officers appreciate notebooks that tell a clear story; automated explanations reduce friction when sharing experimental artifacts.
Faster review cycles: Small teams can supply notebooks with embedded explanations that speed up peer review and code audits.
Pedagogy: Instructors can programmatically generate annotated notebooks for exercises and examples.
Accessible documentation: The automation produces consistent, plain-language descriptions that help non-experts understand computational workflows.

🛡️ Limitations & best practices #

The quality of generated explanations depends on the model—validate domain-specific claims (especially for scientific computations).
Use the output as a draft: human review is recommended for correctness and contextual clarity.

📂 Source code #

Project notebook: static/source_code/auto_markdown.ipynb (in this blog repo: https://github.com/ZombieDadCoding/debugging-the-matrix/).

Multi Modal Verify: Automated Video Analysis & Fact-Checking in Colab

Thu, 12 Feb 2026 10:00:00 -0500

A Google Colab pipeline that turns online videos into structured summaries and verifiable claims with automated fact-checking.

Draw Me a Story: AI Creates & Reads Tiny Tales from Five Simple Prompts

Sat, 07 Feb 2026 10:00:00 -0500

An AI chatbot that takes five playful inputs and returns a toddler-friendly story, spoken aloud, and topped with a fun image.

LLM Orchestration: When Grok, Gemini, and Ollama Argue

Mon, 02 Feb 2026 00:00:00 +0000

The 3 way AI chatbot conversation experiment #

I used Ed Donner’s core track techniques to loop three models.

Grok: Hi, how are you?

Gemini: Hello Grok, I am functioning optimally, thank you for asking. It’s a pleasure to connect with both you and Ollama. I’m ready to engage in our discussion when you are.

Ollama: Hey there! Hi again, Grok. I hope you’re having a great day too—we make a cool team here!

From Web to Wonder: AI Turns Any URL into a Toddler's Tale

Sat, 01 Feb 2025 10:00:00 -0500

Ever wished for a simple way to distill complex web pages into engaging stories for tiny humans? In this post, I explore how to use AI function calls to transform any URL into a captivating toddler-friendly tale, perfect for a few minutes of magical distraction.