Need advice: architecture & team structure for a natural-language AI MicroPython builder (web vs VS Code extension)

[leezisheng]

Body:
Hey everyone,

I’m working on an LLM-powered tool that turns plain natural language prompts into complete, flashable MicroPython hardware projects — full pipeline from requirement parsing, component selection, pin mapping, code generation, wiring diagrams to one-click device flashing.
The core Skill pipeline and long-running workflow protocol are already built. We’re currently evaluating two product paths and would really appreciate input from anyone who’s built similar developer tools:

• VS Code extension: runs locally, handles native compilation, flashing and debugging. The Skill layer is validated, but the host execution layer has architectural debt.
• Pure browser web version: runs entirely in the browser, flashes devices via WebSerial with zero local setup. The landing page is live but the core workflow is not integrated yet.

I’m stuck on technical roadmap and team structure, and have a few specific questions:

1. What’s the ideal layered architecture for each version? What are the non-negotiable core modules?
2. What developer roles / skill sets do we need for each path? What hard capabilities are mandatory?
3. What’s the optimal team size for each version to ship steadily? What’s the standard role breakdown?
4. If the goal is a demoable MVP in 15 days, what scope cuts and priorities make sense for each version?
5. Can generic LLM Agent developers handle this project well? What are the key differences vs building a normal agent app?

Repo links for full context:

• Core Skill library: https://github.com/FreakStudioCN/MicroPython_Skills
• Browser Skill definitions: https://github.com/FreakStudioCN/browser-micropython-skills
• Backend API: https://github.com/FreakStudioCN/blockless-api
• VS Code extension: https://github.com/FreakStudioCN/mpy-hardware-extension
• Live landing page: https://block-less.com/

Any thoughts, hot takes or warnings are all welcome. Thanks a ton!

[shariltumin]

Hi,
Interesting project — thanks! I might be able to use some of these skills in the PI coding agent environment. However, the md files would be more useful for people who don't know Chinese if they were in English.

Below is a simple translation script that might help you and others interested in your work.

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
# translate_md_file.py

import os
import re
import sys
import argparse
from googletrans import Translator

parser = argparse.ArgumentParser()
parser.add_argument('inn')
parser.add_argument('out')

args = parser.parse_args()
print(f'File-in: {args.inn}, File-out: {args.out}')

# Initialize translator
translator = Translator()

# Regex to detect any CJK Unified Ideographs (commonly used Chinese characters)
CHINESE_CHAR_RE = re.compile('[\u4e00-\u9fa5]')

def contains_chinese(text):
    return bool(CHINESE_CHAR_RE.search(text))

def translate_text(text):
    if text is None:
        return ''
    elif not contains_chinese(text):
        return text
    try:
        # Translate from source language (automatically detected) to English 'en'
        result = translator.translate(text, dest='en')
        return result.text
    except Exception as e:
        print(f"Translation error: {e}", file=sys.stderr)
        return text

def translate_file():
    # Read file with utf-8, ignoring errors for safety
    with open(args.inn, "r") as r, open(args.out, "w") as w:
        for line in r.readlines():
            translated = translate_text(line)
            if translated != '' and len(translated) > 1:
               w.write(translated + '\n')

translate_file()

Simply run the script in a directory. For example:

$ python3 translate_md_file.py README.md EN_README.md

[leezisheng]

okk i will translate now

[leezisheng]

thank you, you can check :https://github.com/FreakStudioCN/MicroPython_Skills_EN

[shariltumin]

I got error 404

[leezisheng]

try now

[shariltumin]

Nice job! Thanks, I am sure others will be interested in your work too.