VoxCode/tech_stack.txt at master · Phoenix05420/VoxCode · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
# VoxCode - Tech Stack and Packages

## Language & Environment
* **Python 3.11**
  * Why: Provides the latest performance improvements, native async compatibility, and is the absolute sweet spot for AI dependencies (like `llama-cpp-python` and `js2py`).

## AI Models & Speech Processing
* **Vosk (vosk-model-small)**
  * Why: An extremely fast, lightweight offline ASR (Automatic Speech Recognition) engine. It is used to generate instantaneous, real-time partial text while the user is actively speaking, keeping the application lightning-fast and highly responsive.
* **OpenAI-Whisper (Turbo / Large-v3)**
  * Why: State-of-the-art offline transcription model. It kicks in the massive neural network once the user finishes speaking, replacing the raw Vosk draft with a highly accurate, context-aware final transcript.
* **Llama-CPP-Python (`llama-cpp-python[server]`)**
  * Why: High-performance Python bindings for running quantized LLMs (like `Qwen3.1-1.7B` or `Llama-3`) entirely offline and on CPU/Consumer GPUs. The `[server]` addon hosts the model wrapped inside a FastAPI OpenAI-compatible server on port 8000.
* **SoundDevice**
  * Why: A low-level Python library used to natively interface with the microphone on Windows without compilation issues. It captures raw audio byte streams simultaneously for Vosk and Whisper.

## Web Server & Backend API
* **Flask & Flask-CORS**
  * Why: Very lightweight micro-framework perfect for quickly exposing our `audio`, `generate`, and `snippets` REST API routes on Port 3001 without the bloat of Django.
* **Uvicorn & FastAPI**
  * Why: Under-the-hood ASGI servers used powerfully by `llama-cpp-python` for streaming text tokens chunk-by-chunk over Server-Sent Events (SSE).

## Frontend Architecture & UI
* **Next.js 16 (App Router) & React 19**
  * Why: Provides enterprise-grade React architecture with server-side rendering (SSR), dynamic code splitting, React Suspense boundaries, and built-in API routing rewrites.
* **Monaco Editor (`@monaco-editor/react`)**
  * Why: The exact code editor engine powering VS Code, integrated with custom dark ink styling for our "Light & Tiny Dark" aesthetic.
* **Tailwind CSS v4 & Framer Motion**
  * Why: Delivers crisp utility styling and fluid micro-animations across interactive widgets and voice input visualizers.

## Database & Storage
* **NeonDB PostgreSQL & `psycopg2-binary`**
  * Why: A highly scalable, serverless SQL database used to manage user authentication and the cloud synchronization of their stored code snippets. `psycopg2-binary` is the standard, battle-tested DB adapter to bridge Python and Postgres safely.

## Miscellaneous Extensions
* **ONNXRuntime / PyTorch / NumPy**
  * Why: Tensor computations core to offline speech and language inference models.
* **SpaCy**
  * Why: Advanced Natural Language Processing (NLP) framework, used for potential tokenization or text analysis routines in future iterations of the project workflow.