A.E.G.I.S: A Domain-Specific Large Language Model for Embedded Systems Code Generation via QLoRA Fine-Tuning

A.E.G.I.S (Automated Embedded Generative Intelligence System) is a domain-specific large language model fine-tuned from Qwen2.5-Coder-7B using QLoRA and Unsloth on constrained hardware (Google Colab T4 GPU). Trained on approximately 37,000 Alpaca-format instruction samples spanning Arduino, ESP32, AVR, and STM32 platforms, the model targets embedded systems code generation with platform-aware, memory-constrained output. This paper documents the full pipeline — dataset curation, training configuration, W&B logging, system prompt architecture, and Hugging Face deployment — and evaluates inference quality for real-world embedded development tasks