How I Built an AI Dev Framework as a Coding Student

how-i-built-ai-dev-framework-student-journey

Hey, I'm just a guy in his late forties, buried in textbooks and late-night code sessions. No fancy degree yet, just a burning itch to build stuff that actually works. Last year, I decided to level up to a production ready developer

That's when I stumbled into a rabbit hole of tools: VSCode for editing, Docker for containers, React for slick fronts, Flask for quick backs. But piecing them together? Total mess. Errors everywhere, forgotten setups, and zero way to track what I learned.

I needed a system. Not some off-the-shelf thing, but one born from my own fails and wins. That's how my AI-Powered Development Framework was born.

The Spark: Why I Started Experimenting

It kicked off with a simple goal: build a personal dashboard app. React for the UI, Flask for the API, and Docker to run it without my laptop melting. Sound easy? Nope. First run: VSCode extensions clashing, Docker volumes vanishing, and code mods breaking everything. I spent hours Googling "why is my container ghosting my files?"

Then I found OpenSpec. It's this neat methodology for structuring projects – think proposals for every feature, task lists that actually get done, and specs that evolve as you code. No more vague "add login" notes in a sticky file. But OpenSpec alone felt stiff, like a to-do list without a brain.

Enter Archon, an AI assistant I tinkered with. It's built for research and knowledge tracking, pulling in context from your project to spit out tailored advice. "Hey, how do I secure Flask routes?" Boom – code snippets, best practices, all filed in a growing knowledge base. Using them together? Magic. OpenSpec keeps the plan tight; Archon fills in the gaps with smart research.

My experiments started small:

• Week 1: Docker-ize a basic Flask hello-world. Failed three times before nailing volumes.

• Week 2: Wire React to it via API calls. CORS errors? Archon researched fixes in seconds.

• Week 3: Mod code live in VSCode, sync changes to Git, and pray Docker rebuilds clean.

By month two, I had a Frankenstein setup. It worked for my dashboard, but scaling? Forget it. Time to extract the good bits into a framework.

Building Journal Craft Crew
Crew ai back end, react front end. Reveal soon

How It All Fits Together: The Framework's Core

At its heart, this framework is infrastructure-first. You clone the repo, run a setup script, and boom – a skeleton ready for your project. It's not reinventing the wheel; it's gluing wheels that roll smooth.

OpenSpec: The Backbone for Structure This handles the "what and why." Every feature starts as a change proposal – a Markdown file with sections like problem, solution, tasks, and risks. Tasks break into checkboxes: "[ ] Set up auth routes" becomes atomic steps.

Why it fits: In my experiments, vague plans led to scope creep. OpenSpec forces clarity. For my dashboard, the auth proposal outlined JWT tokens, tasks for Flask routes, and specs for React hooks. No more "I'll figure it out later."

Archon: The AI Brain for Smarts Archon's your co-pilot. It syncs with OpenSpec, so when you mark a task done, it updates the knowledge base. Ask for research? It pulls project context – "Based on your Flask setup, try this Docker compose for multi-service auth."

In practice: During a React state management tweak, I hit a snag with Redux vs. Context. dev-research "react state best practices for flask api" gave me pros/cons, code examples, and even Docker tips for testing. Knowledge sticks around sessions, so next time, it's like I never stopped.

The Integration Layer: Where the Magic Happens This is the glue I scripted in Python. Bidirectional sync keeps OpenSpec tasks mirrored in Archon. Session scripts (dev-start, dev-sync) log everything. VSCode tasks hook in for one-click runs.

For full-stack: Flask backend in /src/api, React in /src/web. Docker-compose spins both, with volumes for hot reloads. Mod code? Git commits trigger Archon to note changes.

Proven in my builds: For the dashboard, sync handled 50+ tasks across 10 sessions. Zero lost context, and dev speed jumped 30% after week four.

Step-by-Step: Building Your First App with It

Want to try? Here's how I did my dashboard, framework-ified.

1. Clone and Init

   text

   git clone https://github.com/RegardV/ai-powered-development-framework.git my-dashboardcd my-dashboard./setup-framework.sh./init-project.sh "Personal Dashboard"

   Sets up folders: /openspec for plans, /src for code, .env for configs.

2. Plan a Feature

   text

   source ./dev-workflow.shdev-startdev-plan "user login system"

   Creates /openspec/changes/20241105-user-login/ with proposal.md and tasks.md.

3. Research and Build

   text

   dev-research "flask jwt auth with react"

   AI drops: Code for Flask routes, React login form, Docker secrets. I copied, tweaked in VSCode.

4. Code and Sync

   • Flask: Add /src/api/auth.py with JWT logic.

   • React: /src/web/components/Login.js with axios calls.

   • Docker: Update compose.yml for postgres db.

   text

   dev-syncdev-end

5. Deploy Local Docker up, browse localhost:3000. Login works. Archon logs: "Task complete, knowledge added: JWT expiry handling."

That's one feature. Repeat for dashboards, charts – boom, app done.

Challenges I Faced (And Fixed)

Honest talk: Not all smooth. Early Docker setups ate RAM; fixed with compose limits. Archon sync lagged? Added interval checks in scripts. React-Flask CORS? OpenSpec spec forced me to doc it upfront.

As a student, time was tight. Framework cut setup from 2 hours to 10 minutes per project. Now I experiment faster –

Where It Fits in the Bigger Picture

This isn't enterprise-only. It's for students like me, juggling classes and side hustles. Fits indie devs building MVPs, or teams wanting AI without bloat. Compared to cookie-cutter boilers? This learns with you – Archon grows the KB, OpenSpec tracks evo.

Future? Adding CI/CD hooks, more AI models. Fork it, tweak for your stack.

Since this blog was created I discovered how to functional orchestrate the build process.
See this blog post.

Blogpost soon: SO HOW DO WE DO THIS BETTER
https://github.com/RegardV/Buider-Craft-Crew

But on to the next Experiment:

If you're tinkering, start small. Your experiments might birth the next framework too.

(Last updated: November 2025. Based on real student builds.)