Remove demo script from public repository

.gitignore

@@ -25,3 +25,4 @@ __pycache__/
 # OS
 .DS_Store
 Thumbs.db
+DEMO_SCRIPT.md

DEMO_SCRIPT.md

@@ -1,88 +0,0 @@
-# 🚀 ROCm Forge: The Ultimate Pitch & Live Demo Script
-
-*This script is designed to be conversational, confident, and highly impactful. It directly addresses the judges' technical backgrounds while keeping the narrative extremely simple and easy to follow.*
-
----
-
-## 🎙️ PART 1: The Hook & The "Why Us?" (0:00 – 1:00)
-
-**[Screen is showing the dark, sleek ROCm Forge UI]**
-
-**You:** 
-"Hi everyone, we are Team Cipher, and this is ROCm Forge. 
-
-To migrate code from NVIDIA to AMD, you might think it's as simple as finding the word 'cuda' and replacing it with 'hip'. I guarantee you, several teams today built exactly that—a basic regex 'Find and Replace' tool. 
-
-Here is why ROCm Forge is entirely different: **We built a compiler-level, 9-Agent AI Architecture.** 
-
-Other tools just read text. ROCm Forge actually understands the physical hardware differences between an NVIDIA H100 and an AMD MI300X. It predicts build errors *before* you compile. It flags silent mathematical failures. We aren't just changing words; we are doing deep architectural translation.
-
-Let me show you exactly what I mean."
-
----
-
-## 💻 PART 2: The Live Demo (1:00 – 3:00)
-
-### Click 1: PyTorch ResNet (The Standard Case)
-**[Select "PyTorch ResNet Training" → Click MIGRATE]**
-
-**You:**
-"Let’s start simple. A standard PyTorch training script. We click migrate."
-
-**[Point to the green 95% AMD Readiness Score]**
-"Our 9 agents fire off instantly. Look at the dashboard: AMD Readiness is 95%. Everything is green. The AI knows that PyTorch's `.cuda()` API actually works transparently on ROCm, so it marks this as low-risk and gives us the green light to deploy. For simple scripts, it's an easy win."
-
----
-
-### Click 2: Hugging Face & vLLM (The AI Ecosystem)
-**[Select "Hugging Face Fine-Tuning" → Click MIGRATE]**
-*(Wait 2 seconds)*
-**[Select "LLM Inference with vLLM" → Click MIGRATE]**
-
-**You:**
-"What about modern generative AI? If we load a Hugging Face QLoRA script or a vLLM server... same result. The AI intelligently swaps NVIDIA wheels for ROCm wheels, handles `bitsandbytes`, and changes environment variables like `CUDA_VISIBLE_DEVICES` to `HIP_VISIBLE_DEVICES`. High readiness, ready to run."
-
----
-
-### Click 3: Dockerfile (Infrastructure)
-**[Select "Dockerfile (NVIDIA CUDA Base)" → Click MIGRATE]**
-
-**You:**
-"It even handles infrastructure. If I give it a production Dockerfile, the Deployer Agent automatically rips out the NVIDIA base images and injects the exact ROCm 6.2 Ubuntu images, removing obsolete NVIDIA driver flags along the way."
-
----
-
-### Click 4: The Final Boss (The Differentiator)
-**[Select "Tensor Core WMMA Kernel" → Click MIGRATE]**
-
-**You:**
-"But here is where we separate ourselves from everyone else. I’m going to feed it the 'Final Boss'—a low-level C++ Tensor Core kernel using NVIDIA WMMA intrinsics. A normal Find-and-Replace tool would swap the headers and call it a day... which would result in a broken, compiling nightmare."
-
-**[Point to the dashboard turning RED and 0%]**
-"Look at ROCm Forge. The Readiness Score plunges to ZERO. The Risk Heatmap is glowing red. The engine is literally screaming at us."
-
-**[Click on the "Agent Trace" Tab]**
-"And here is why: Our Hardware-Aware Scanner caught something critical. It found a hardcoded warp size of 32. If we just deployed this on an AMD GPU, which uses a 64-wide wavefront, it would silently compute the wrong math. 
-
-Our Build Error Copilot goes even further, warning us to link the `-lrocblas` library to avoid a compiler crash. *This* is the power of a 9-Agent architecture."
-
----
-
-## 🎯 PART 3: The Close (3:00 – 3:30)
-
-**[Switch back to the main UI]**
-
-**You:**
-"ROCm Forge doesn’t just migrate code. It protects engineers from days of debugging hardware-level compiler errors. 
-
-We’ve packaged an enterprise-grade migration consultant into a sleek, instant AI tool. Thank you, and we’d love to answer your questions."
-
----
-
-## 🧠 Cheat Sheet: If Judges ask "What are the 9 Agents?"
-If they ask, confidently list a few of the coolest ones to show off your backend engineering:
-1. **Hardware-Aware Scanner:** Looks for implicit hardware assumptions (like Warp Size 32 vs 64).
-2. **Build Error Copilot:** Cross-references code against a runbook of common ROCm compiler errors to fix them before compilation.
-3. **Health Monitor:** Calculates the AMD Readiness score and drift detection.
-4. **Deployer Agent:** Specifically handles Docker, dependencies, and environment setup.
-5. **AST Refactorer:** Does semantic syntax-tree replacement, not just regex.