Hi everyone,
I've been hearing the term 'vibe coding' recently, and somehow I found myself riding that wave. After a few days of just tinkering with an AI, I managed to build and even publish two small apps to the Microsoft Store. It's been a truly amazing and surprising experience!
Riding that high, I got ambitious and started on a third program, but I've hit an unexpected wall and would love to get your advice.
The Tools I'm Building
My goal is to make the PC experience smarter using voice commands and AI.
- OmniVoiceKey: A free voice recognition utility I built out of frustration with the native voice recognition in Windows 10. It lets you use a hotkey to recognize your voice and preview the text, and it works directly in the Windows environment without needing any special APIs.
- QuickAskAI: A utility that lets you instantly ask an AI a question with a hotkey and get a quick answer.
(Both of these are already on the Microsoft Store!)
And now, I'm working on a 'personal search engine' called ContextIndexer to tie them all together. It's a system that automatically saves my voice commands and AI search history locally, allowing the AI to reference these logs for smarter, context-aware answers in the future.
My Motivation: Solving My Own Problem
Honestly, this search engine project started because I was trying to solve a problem I was personally facing while working with AIs.
I rely heavily on AI for development, but I found myself having to create a new chat session more than 10 times a day because I thought longer chats cost more in tokens as the AI re-reads the whole context. As a result, the conversation context was constantly being lost, which was incredibly frustrating. On top of that, the token costs for re-explaining the context were growing exponentially—I spent over $50 in just three days.
So, I decided, "Why not build a database of all my development logs and let the AI search it when needed?" And that's how the ContextIndexer project was born.
The irony is, while building this very system meant to save costs, I've run into a technical roadblock.
The Current Problem: Python Virtual Environment
The core logic of this search system is being developed in Python, but for some reason, my Python 3.12 virtual environment isn't working correctly, and it's blocking my progress.
- OS: Windows 10
-
Execution Method: An
index.batbatch file that directly calls thepython.exefrom inside the.venvfolder.
Even though I'm sure I created the .venv with Python 3.12, running the index.bat file keeps calling a Python 3.14 version, which causes an error.
(Most Importantly!) My Debugging Journey So Far:
Here is a log of the steps I've taken to try and solve this on my own:
text
# Python Environment Debug Log (2025-12-04)
## Final Goal
Get the ContextIndexer project to run correctly in a stable Python 3.12 environment.
## Troubleshooting Timeline
### Attempt 1: Modify System PATH
- Hypothesis: The system's default Python version was incorrectly set to 3.14.
- Action: Elevated the priority of the Python 3.12 path in the system PATH variables. Confirmed `python --version` showed 3.12 in a new terminal.
- Result: Failure. Running index.bat led to other errors, like not finding `pip`.
- Analysis: Discovered that index.bat references the project's virtual environment (.venv) regardless of the system PATH.
### Attempt 2: Recreate Virtual Environment (.venv) - 1st time
- Hypothesis: The existing .venv might be corrupted.
- Action: Created a fresh venv using `python -m venv .venv` and installed libraries.
- Result: Failure. `ModuleNotFoundError`.
### Attempt 3: Resolving Dependencies (Dependency Hell)
- Hypothesis: Version conflicts between libraries, likely `langchain-google-genai` and `google-generativeai`.
- Action: Tried reinstalling and realigning packages.
- Result: The installation itself succeeded, but the root problem remained.
### Attempt 4: Final Test and Root Cause Discovery
- Action: Ran `.\index.bat` again.
- Result: Critical Failure. `TypeError: Metaclasses with custom tp_new are not supported.`
- Final Conclusion: I finally noticed that the error log path referenced `Python314`. This means the command I used to create the venv (`python -m venv .venv`) was itself being run by the system's Python 3.14, not the intended 3.12, creating a contaminated virtual environment from the start.
Top comments (0)