psyxd-nepali-jarvis

PsyxD Nepali Jarvis

A multilingual Nepali AI assistant powered by fine-tuned Qwen3-8B.

Features

• Multilingual: Speaks Nepali, English, and Hindi naturally
• Coding Expert: Explains programming concepts in Nepali + English
• Tool Calling: File management, browser automation, Python execution, GitHub, smart-home, charts
• Persistent Memory: Remembers conversations using embeddings + vector store
• Voice I/O: Speech-to-text and text-to-speech support
• Local & Cloud: Deployable locally or on Hugging Face

Project Structure

├── app/agent/          Agent planner, router, executor
├── app/tools/          Tool integrations
├── app/memory/         Persistent memory with vector store
├── app/voice/          Speech-to-text & text-to-speech
├── app/personality/    System prompt & cultural traits
├── app/ui/             Interface
├── datasets/           Training datasets (JSONL)
├── training/           Training, merging, evaluation scripts
├── configs/            Configuration files

Setup

python -m venv .venv
# Windows
.venv\Scripts\activate
# Linux/Mac
source .venv/bin/activate

pip install -r requirements.txt

Copy .env.example to .env and fill in your tokens.

Training

python training/train.py

Run

uvicorn app.main:app --reload

Push to Hugging Face

from transformers import AutoModelForCausalLM, AutoTokenizer

model.push_to_hub("psyflowz/psyxd-nepali-jarvis")
tokenizer.push_to_hub("psyflowz/psyxd-nepali-jarvis")

License

MIT