AM_Eng-Web

AM_Eng is a full-stack AI chat application built with Django and powered by a fine-tuned Large Language Model (LLM). The system delivers a ChatGPT-like experience with a custom-trained LoRA model, enabling contextual, technical, and concise conversations focused on computer science and engineering topics.

Core Idea

The goal of AM_Eng is to explore how custom-trained LLMs can be embedded into real-world applications, focusing on:

• Context-aware technical assistance
• Persistent memory across interactions
• Efficient inference on limited hardware
• Clean and responsive user experience

Key Features

• ChatGPT-like interface with real-time interaction
• Custom LoRA fine-tuned model for technical domains
• Persistent memory system (user fact extraction + reuse)
• Optimized inference pipeline using 4-bit quantization
• Secure Markdown rendering (XSS-safe with DOMPurify)
• Context-aware responses using conversation history

Backend

• Django (Python)
• REST-like endpoint for LLM interaction
• Session-based conversation handling

AI Layer

• Base Model: Qwen2.5-3B-Instruct
• Fine-tuning: LoRA (PEFT)
• Inference: Transformers + BitsAndBytes (4-bit)

Model

Fine-tuned LoRA model available at: https://huggingface.co/PGFerraz/qwen-alpaca20k-amEng

Highlights:

• Reduced VRAM usage via quantization
• Domain-adapted responses for engineering contexts
• Modular integration with base model

Setup

Prerequisites

• Python 3.10+
• Git
• (Recommended) NVIDIA GPU with CUDA
• pip / virtualenv

Optional:

• CUDA Toolkit
• nvidia-smi working

---

1. Clone

git clone https://github.com/PGFerraz/AM_Eng-Web.git
cd AM_Eng-Web

2. Virtual Enviroment

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

# Windows
.venv\Scripts\activate

3. Install Dependencies

pip install --upgrade pip
pip install -r requirements.txt

4. Model

The model is downloaded automatically from Hugging Face: https://huggingface.co/PGFerraz/qwen-alpaca20k-amEng First run may take time.

4. Run

python manage.py runserver

Open: http://127.0.0.1:8000