This fun personal project i started with provided me tons of benefits, I call it Bite-Size Notes! 🍪

It started off as an end user, my initial thoughts was how can i maximize my time as i absorb knowledge from tutorial videos or live conferences without missing the important details?

The solution was a personal notetaker that can listen alongside, take notes and summarize it for me. Whether i am participating at a live event or live only stream, the assistant is readily available to note everything down.

I proceeded to blueprint and spec the architecture with my rocket book on what i need to achieve this product.

Data is very important to me so i took the local-first approach, but of course model provider agnostic option being available. Windows first Mac later.

The full python only stack: Summarization Model - Qwen3-4b-GGUF (Local) Transcription Model - Whisper GGUF (Local) PySide6 - GUI Faster-Whisper - Transcribe engine Llama-cpp - LLM engine Huggingface - Model downloading json - Session files stored in computer

The MVP in a nutshell

1. Start a new session
2. Start recording
3. Whisper captures transcripts live while you take notes on the side
4. Qwen captures array of transcript and your notes then applies a system prompt to get a deterministic output tailored for meetings
5. Export notes as txt, profit.
6. Retains with json files stored in your local appdata/roaming to represent sessions
7. Ability to download the transcripts if you don't like your local summarization model

First customer was my girlfriend and bias or not she loved it, as she can use it for school or work meetings to focus on what matters most, the lectures and dialogs.

No scale needed, the app sits in your computer but is limited to your hardware specs of course, unless you decide to refactor to use open ai spec.

Anyway i solved my problem with automating notetaking at conferences and tutorials; it was a fun learning experience. Shipping as a executable is the next step!

Reference project: https://lnkd.in/gz9S9vka

hashtag#Learning hashtag#SpeechToText hashtag#Summarization hashtag#LocalFirst hashtag#AutomateMyLife hashtag#AI

I have been addicted to using this coding assistant tool for app development.

Claude Code - not just your usual coding assistant, but an agentic coding tool that integrates with your terminal and IDE to enhance productivity, whether it is debugging and refactoring or testing and prototyping.

What i love most is taking a different perspective in development where i start to wear the hat of an architect/end user, to articulate system designs in a back and forth consultation with Claude Code. The rigorous planning with the coding assistant provided me the most accurate design that i needed for my product. The assistant always lays out the plan allowing edits before i pull the trigger to execute build.

One amazing example i used claude code for, was building my personal GitHub page. I was amazed at how opus 4.6 understood my requirements so well with help of skills.md and rolled out a quick shopify-esque MVP site, it's beautiful. I will share the site below.

The key takeaways:

• Always plan what you need in your project, one turn execution usually results in bad product
• Skill.md are a godsend when it comes to extending expertise outside the coding assistant's knowledge when needed; progressive disclosure
• Clear context when you need a brand new task, the coding assistant can only carry so much in the context window before degrading
• The quality in Gen AI output is only as good as how you articulate your ideas; garbage in garbage out
• Knowing the fundamentals and tools in your field's system design patterns often provide the best product rather than open ended questions
• Zero trust, always be skeptical and read what the assistant will output or take control; the human in the loop

Next learning phase, piping apps on free tiered Vercel with claude code!

MVP: https://pcphil.github.io/

hashtag#ClaudeCode hashtag#CodingAssistant hashtag#AgenticWorkflow hashtag#Learning

Welcome to my blog! This marks the day, i set up the github page!