Startup Advisor AI

is a domain-specific large language model designed to act as an expert consultant for startups . Developed for the AMD Developer Hackathon, the project solves the problem of generic, "wiki-style" advice generated by standard AI models by delivering immediate, high-impact guidance . The model was built by distilling core principles from renowned Silicon Valley business literature (such as The Lean Startup, Zero to One, and Blitzscaling) into a meticulously engineered dataset of 1,564 ChatML instruction-following pairs . To ensure tactical execution, the model adheres to a strict 4-pillar response architecture: Direct Actionable Advice, Why This Matters, Real-World Examples (grounded in 2025 context), and 'Avoid' variables . Built on the Qwen2.5-7B-Instruct base model, the fine-tuning process was hyper-efficient. By leveraging Unsloth optimization and QLoRA 4-bit quantization, the 7-billion parameter model was trained in under 30 minutes using only 5.5 GB of VRAM . Crucially, the model architecture and LoRA adapters are 100% compatible with AMD's ROCm ecosystem (including AMD Instinct™ GPUs and Ryzen™ AI-powered PCs) for high-performance production inference . The project is accessible via a premium, fully interactive Gradio chat interface deployed on Hugging Face Spaces