KentLikesTech.com - computer nerd, GrandPa, and wanna be hacker

This text is served by GitHub Pages.

Effort from 10.9.2026 & 10.10.2026

h2>

No nerd stuff. 100% dedicated to family. Swiming, bike rides, painting cooking, hide and seek, gaming, BBQ, and just good vibes with the crew.

Rise and fn grind Monday, will be a little sus the next 2-3 days with family opportunties, but so worth it

Effort from today 10.8.2026

Very little coding today, spent most of the day focusing on tools

Utilized the Google AI Edge Foresite, this tool has a lot of promise and plan to use on a webinar I have this weekend.

Played with Grok Imagine, fun visual tool not extremely relevant to my workflow.

Utilized my RAG and created 50 item quizes for SecAI+.

Downloaded GitHub Copilot and will test that in my work flow.

Hit 80 days straight on Tryhackme.com - Currently at 49,613 Points

Effort from today 10.7.2026

Spent the majority of today on my windows laptop playing with the NPU and learning how to take advantage of that hardware

Set up Python 3.11 ARM64 and installed ONNX Runtime’s QNN package and dependencies.

Configured the benchmark to use Qualcomm’s QNN execution provider and disabled CPU fallback so NPU runs cannot be silently counted as CPU runs.

Updated the model export and benchmark scripts to support a fixed batch size of 32.

Compared CPU and NPU performance on a tiny linear model, then switched to ResNet-18 for a more representative benchmark.

ResNet-18 benchmark: Input shape was 1 × 3 × 224 × 224, with CPU fallback disabled for the QNN run.

CPU: 17.02 ms per image; 58.77 images per second.

Qualcomm NPU (QNN): 1.55 ms per image; 647.15 images per second.

Result: The NPU was approximately 11× faster than the CPU in this test.

My System

System Specifications

Operating system: Windows 11 Pro, ARM64, build 26200

CPU: Qualcomm Snapdragon X Elite (X1E80100), 12 cores, up to 3.40 GHz

Memory: 63.5 GB usable RAM

GPU: Qualcomm Adreno X1-85, driver version 31.0.133.1

NPU: Qualcomm Hexagon NPU, driver version 30.0.220.3000

Benchmark environment: Python 3.11.9 ARM64 with ONNX Runtime QNN

Effort from today 10.6.2026

This morning kinda sucked after breakfast and coffee

Hit the dreaded... Individual quota limit hit after basically one night of heavy coding...

Briefly played with Google AI Studio... Not enough to know value added for my efforts

Downloaded and fed into my RAG additiona CompTIA SecAI+ documentation

Liked and commented on all of my familys TikTok videos

Uploaded my AI cheat sheet to TikTok https://www.tiktok.com/@kentlikestech/photo/7693011281854467359

Today, I got the local AI and transcription work moving:

Configured the RAG GUI to use the locally installed mistral:latest model after diagnosing the invalid Ollama model tag.

Troubleshot the WebM transcription pipeline: found the correct input file, fixed the PyAV incompatibility by installing PyAV 18.1.0 in the virtual environment, and got the transcript generated. transcription.txt is present and contains about 16 KB of text.

Investigated Python SSL and dependency issues, including a missing certificate bundle and package files unavailable in a OneDrive-synced environment. The durable environment fix is to keep the virtual environment outside OneDrive.

The main win: the transcription moved from crashing on setup to producing a transcript, and the RAG app is configured for Mistral.

Effort from today 10.5.2026

Super hard workout today. Startted using bands along with my s 2 s time to work on chest and my fat tummy

Downloaded and installed Antigravity by Google, IDE looks really close to VSCode, linked to my OneDrive Pyhthon folder and ran my Nvidia chatbot like butter

Does not look like the learning curve will be bad if I decide to switch... Will link to my GitHub later and see how that integrates

I use local models (ollama) for a lot of my RAG stuff so need to see how easy it will be to integrate this in the new IDE

Did a couple quizes today since my VM kept shitting the bed from the gym... Painful

Thinking about going back to my Windows ARM laptop for a little bit (maybe) of a speed boost... I currently use a Mac and will miss not having FaceTime and Imessages on screen

Heavy Lifting, all work completed in Antigravity

Transformed a basic CLI prototype into an enterprise-grade, high-speed personal Second Brain RAG system capable of parsing multi-format documents (PDFs, Word docs, Markdown, text), executing sub-second incremental vector syncing with Ollama & ChromaDB, and delivering real-time streaming answers through a custom glassmorphic Web UI with interactive source citations.

And the best part!

- **100% Air-Gapped & Private:** Powered entirely by local Ollama models (`llama3.1:8b` and `nomic-embed-text`) with zero data leaving the local machine.

Effort from today 10.4.2026

Today was my offical day off! Did not get on my laptop until tonight! Had a wonderful breakfast with my grand daughter then a swim day with the ladies! This weeks focus is workout and TryHackMe so may not get as much coding complete!

Effort from today 10.3.2026

Consolidated all my random Python code onto GitHub for a central repository for my code

Working to learn more about the GitHub process and how I can integrate my coding effort inside GitHub instead of the circle jerk of uploading

Got to get back to my effort on TryHackMe.com currently at 74 days in a row but only 2 streak freezes left

Effort from today 10.2.2026

gemini_script.py

Basic function: an interactive chat script for Google Gemini. It reads your GEMINI_API_KEY environment variable, asks you a question in a loop, sends it to the Gemini API, and prints the response until you type exit or quit.

gui youtube downloader.py

Basic function: a Tkinter-based YouTube downloader. It gives you a small GUI with a URL field, a folder picker, and a Download button. It uses yt-dlp to download the selected video/audio into the chosen folder and shows success or error popups.

send_email.py

Basic function: an email-sending helper for SMTP Gmail. It builds an HTML email message, sets the subject and recipient, connects to Gmail via SSL, logs in, and sends the message. It looks like it is intended for a daily CVE/briefing email.

Please let me know if you have any suggestions.