Techalicious ~ GTA's Tech Hobbyist Social Group

Ever wanted an AI that actually gets you? Not some corporate assistant that sounds like it was raised by a customer service manual, but a genuine digital companion with personality, opinions, and maybe even a sense of humor?

This week we're building custom AI chatbots from scratch using free, local tools. No cloud subscriptions. No sending your conversations to Big Tech. Just you, your computer, and a new digital friend.

🤖 What we'll cover:

• Running AI models on your own machine (goodbye subscription fees)
• The secret sauce: character cards that actually work
• Scene-based prompting and why it beats instruction lists every time
• Making your bot sound human and killing those cringy AI-isms
• Giving your companion a real voice, not just a list of traits

Whether you want a study buddy, a creative collaborator, a sounding board for ideas, or just someone to chat with who won't judge your 2am snack choices, we'll show you how to make it happen.

🤖 You'll leave with:

• A working AI companion running on your own hardware
• A template for building more personas
• The know-how to tweak personality, tone, and conversation style

No coding experience required!

Ever wanted to just describe what you need and have working code appear? That's vibe coding, and Qwen Code brings it right to your command line.

Qwen Code is Alibaba's open-source CLI for agentic coding. By default it calls their cloud API, but since it uses the OpenAI-compatible format, you can point it at your local Ollama instance instead. All the convenience of a polished CLI tool, none of the cloud dependency.

🧙‍♂️ What we'll cover:

• Installing Qwen Code (just needs Node.js 20+)
• Running Qwen3-Coder locally through Ollama
• Pointing the CLI at localhost instead of the cloud
• Basic workflow for generating, refactoring, and explaining code
• When vibe coding saves time vs when to just write it yourself

🧙‍♂️ Who's this for?

Anyone comfortable with the terminal who wants AI-assisted coding without sending their code to someone else's servers.

No cloud subscriptions required. Local AI doing useful things.

Pattern matching is one of those skills that separates "I can code" from "I can wrangle anything." This session goes beyond the basics into the powerful features most people don't know exist.

🧠 What We're Covering:

• The (?x) extended modifier that lets you write regex with comments and whitespace. Finally, patterns that humans can actually read.
• Regex subroutines using (?&name) so you can define a pattern once and reuse it. DRY principle applied to regex.
• Lookaheads and lookbehinds for zero-width assertions. Check for something without consuming it.
• Non-greedy quantifiers when you want the shortest match, not the longest.
• Named capture groups with (?...) for readable, maintainable extractions.
• Conditional patterns with (?(condition)yes|no) for context-aware matching.

🧠 Why This Matters for AI Work:

If you're running local LLMs through Ollama or OpenWebUI, you're dealing with text. Lots of it. Model outputs wrapped in markdown code fences. Log files full of timestamps and metrics. User requests that need parsing into structured prompts.

We'll work through real examples:

• Extracting JSON from chatty LLM responses (they love to add "Sure! Here's the data...")
• Parsing Ollama logs for debugging and monitoring
• Cleaning AI-generated text before downstream processing
• Building dynamic prompts from captured user input

🧠 Who Should Join:

Anyone comfortable with basic regex who wants to level up. If you know .* and [a-z]+ but (?x) and (?&name) look like hieroglyphics, this session is for you.