Add read me

README.md

@@ -6,6 +6,8 @@ colorTo: indigo
 sdk: gradio
 sdk_version: 6.18.0
 app_file: app.py
+tags:
+  - build-small-hackathon
 pinned: false
 ---
 
@@ -45,6 +47,7 @@ git commit-at
 - ✅ Ticket number integration
 - ✅ Live branch graph visualization
 - ✅ Commit history tracking
+- 🧠 Fine-tuned model integration (see below)
 
 ## Directory Structure
 
@@ -95,6 +98,46 @@ git commit-at
 # - Confirm and commit
 ```
 
+## Fine-Tuned Model Integration
+
+`git-commit-at` now uses a **custom fine-tuned model** as its primary AI, with automatic fallback to a general model.
+
+### Model
+
+**[Bhargavi5q1/git-commit-messages](https://huggingface.co/Bhargavi5q1/git-commit-messages)** — Fine-tuned on 2,179 real git diffs using Unsloth + QLoRA on a Qwen2.5-Coder 1.5B base model. Trained to output conventional commit messages in one line with no explanation.
+
+### How it works
+
+```
+git diff (staged changes)
+        │
+        ▼
+┌─────────────────────────────────┐
+│  Fine-tuned model (primary)     │  hf.co/Bhargavi5q1/git-commit-messages
+│  via Ollama                     │
+└────────────┬────────────────────┘
+             │ fails?
+             ▼
+┌─────────────────────────────────┐
+│  qwen2.5-coder:1.5b (fallback)  │  general-purpose model
+└─────────────────────────────────┘
+```
+
+- Primary model is pulled automatically on first run (~986 MB)
+- If unavailable, falls back silently to `qwen2.5-coder:1.5b`
+- Terminal shows which model was used on each run
+
+### Training details
+
+| | |
+|---|---|
+| Base model | Qwen2.5-Coder 1.5B |
+| Method | QLoRA (Unsloth) |
+| Dataset | 2,179 git diff → commit message pairs |
+| Epochs | 3 |
+| Format | Conventional Commits |
+| Export | GGUF Q4_K_M (~986 MB) |
+
 ## Architecture
 
 ```
@@ -230,8 +273,8 @@ node wrapper.js
 **Q: Do I need Ollama installed locally?**
 A: No! Ollama runs in a Docker container. Just ensure Docker is running.
 
-**Q: Can I use a different AI model?**
-A: Yes! Edit the `MODEL` variable in `git-commit-at/index.js`.
+**Q: Which AI model does this use? (Inprogress)**
+A: A custom fine-tuned model ([Bhargavi5q1/git-commit-messages](https://huggingface.co/Bhargavi5q1/git-commit-messages)) trained specifically for git commit messages. Falls back to `qwen2.5-coder:1.5b` if unavailable.
 
 **Q: Does this work offline?**
 A: After the first run (which downloads the AI model), yes!
@@ -249,7 +292,7 @@ MIT — See [LICENSE](LICENSE) for details
 ## Links
 
 - 🐳 Docker Hub: [git-commit-at](https://hub.docker.com)
-- 🤗 Hugging Face: [Spaces](https://huggingface.co/spaces/build-small-hackathon/git-commit-a)
+- 🤗 Hugging Face: [Fine-tuned model](https://huggingface.co/Bhargavi5q1/git-commit-messages) · [Spaces](https://huggingface.co/spaces/build-small-hackathon/git-commit-a)
 - 📦 NPM: [git-commit-at](https://npmjs.com/package/git-commit-at)
 
 ## Support