ai-code-reviewer

GitLab AI Code Reviewer

GitLab Demo

🖼 Architecture Flow

flowchart TD
    A[Developer opens Merge Request] --> B[CI triggered]
    B --> C[Run AI Review Job]
    C --> D[Flask API: Receives MR ID & Project ID]
    D --> E[Fetch diff & description from GitLab API]
    E --> F[Send to OpenAI GPT-4 API]
    F --> G[Result: pass/fail + comments]
    G --> H[Post comments to GitLab MR]
    G --> I[Log result into Postgres]
    I --> J[Dashboard UI with Chart.js]

CI → FastAPI → GPT → GitLab comment + DB log → Dashboard

✨ Features

• ✅ Auto-fetch MR diff & description
• 🤖 GPT-4-based code review [OpenAI / Github Models]
• 💬 Inline comments to GitLab MR
• 🧩 GitLab CI/CD ready
• 🏷️ Adds badge label ai-reviewed:pass|fail
• 📥 Logs feedback rating (1–5) to DB
• 🛠️ Optional: Auto-fix patch block suggestion

📦 Project Structure

├── app/
│   ├── api.py 
│   ├── main.py                # Main script for review flow
├── config/                 # Static files and charts
├── docker/                   
│   ├── Dockerfile             # Dockerized deployment
│   └── nginx.conf             # Proxy for Flask
├── fine-tune/                
│   ├── train_code.py          # Fine-tune CodeT5/StarCoder (optional)
│   └── train.json             # Training samples
├── infra/
│   ├── .gitlab-ci.yml         # GitLab CI job integration
│   └── docker-compose.yml     # Local/dev deployment
└── README.md                  # Project overview

⚙️ Usage

1. Install dependencies

pip install -r api/requirements.txt

2. Set environment variables

export GITLAB_TOKEN=your_token
export OPENAI_API_KEY=your_openai_key
export CI_PROJECT_ID=your_project_id
export CI_MERGE_REQUEST_IID=your_mr_iid

3. Run the review script

python api/main.py

This will:

• Fetch diff and description
• Send to OpenAI
• Post comments to GitLab
• Log to database

🚀 GitLab CI Integration

Example .gitlab-ci.yml job:

stages:
  - ai_review

auto_review:
  stage: ai_review
  image: python:3.10
  script:
    - pip install -r api/requirements.txt
    - python api/main.py
  only:
    - merge_requests

📊 Dashboard (Optional)

Flask + Jinja2 + Chart.js based UI

Features:

• MR history
• PASS/FAIL filter
• Comments per MR
• Excel/PDF export
• Submit feedback rating

Feedback Form:

<form method="POST">
  <label for="rating">Was the AI suggestion helpful?</label>
  <select name="rating">
    <option value="5">Excellent</option>
    <option value="4">Good</option>
    <option value="3">Average</option>
    <option value="2">Poor</option>
    <option value="1">Wrong</option>
  </select>
  <button type="submit">Submit</button>
</form>

Useful for building fine-tuning dataset

🧠 Advanced Extensions

🛠 Auto-fix Patch Suggestion

Append this to your prompt:

If fixable, include a "fix" field:
--- path/to/file.ext
+++ path/to/file.ext
@@ -42,7 +42,7 @@
- old code
+ new code

Use GitLab markdown to render in MR comment.

⭐ Feedback Rating API

@app.route("/feedback", methods=["GET", "POST"])
def feedback():
    if request.method == "POST":
        rating = int(request.form["rating"])
        mr_id = request.form["mr_id"])
        cur = conn.cursor()
        cur.execute("UPDATE ai_reviews SET feedback_rating = %s WHERE mr_iid = %s", (rating, mr_id))
        conn.commit()
        return "Thank you!"
    return render_template("feedback_form.html")

🏷 Add GitLab MR Label

def add_label(label):
    url = f"{GITLAB_API}/projects/{PROJECT_ID}/merge_requests/{MR_IID}/labels"
    headers = {"PRIVATE-TOKEN": GITLAB_TOKEN}
    payload = {"labels": label}
    requests.put(url, headers=headers, json=payload)

# Usage:
add_label("ai-reviewed:pass")

📌 Notes

• Ensure GitLab token has api scope
• OpenAI usage may incur cost
• Prompt/model can be adjusted to suit context

Feel free to fork, contribute, or open issues!