🎯 PitStop AI — PR Risk Analysis System

PitStop AI is a high-performance, privacy-first risk analysis system for Bitbucket Pull Requests. It leverages Gemini 2.0 and custom ML heuristics to provide explainable risk scores, actionable suggestions, and historical context—all without your code ever leaving the Atlassian Forge environment.

---

✨ Key Features

• 🚀 Dual-Engine Analysis: Combines Gemini AI logic with a TF-IDF + Cosine Similarity ML engine.
• 🛡️ Structural Protection: Automatically flags breaking changes in critical files (Auth, API, Config).
• 📉 Documentation Discount: Intelligent enough to recognize low-risk documentation-only changes.
• 🔄 Stale-Free Recommendations: Content-aware caching ensures you always see suggestions for your latest code version.
• 🏠 Secure by Design: Runs entirely on Runs on Atlassian (No external network egress required).

---

🏗️ System Architecture

PitStop AI operates as an integrated Forge application within the Bitbucket ecosystem.

graph TB
    subgraph "Bitbucket Cloud"
        PR[Pull Request]
        BB_API[Bitbucket API]
    end
    
    subgraph "Atlassian Forge Runtime"
        subgraph "Backend (Node.js 20)"
            Resolver[Forge Resolver]
            ML[ML Service v3<br/>TF-IDF + Cosine]
            Gemini[Gemini AI Service]
            Storage[(Forge Storage)]
        end
        
        subgraph "Frontend (React)"
            UI[React UI Panel]
            Components[UI Components]
        end
    end
    
    PR -->|Open PR| BB_API
    BB_API -->|Request Panel| Resolver
    Resolver --> ML
    Resolver --> Gemini
    ML --> Storage
    Gemini -->|Telemetry| ObservAI[ObservAI SDK]
    Resolver -->|Analysis Results| UI
    UI --> Components
    Components -->|Rendered Panel| PR

Loading

---

🔬 How it Works: The ML Pipeline

We don't just guess risk; we calculate it using a multi-dimensional approach.

flowchart LR
    subgraph "Input"
        PR_Text[PR Title + Description]
        PR_Meta[Metadata additions, deletions, files]
    end
    
    subgraph "Processing"
        TF_IDF[TF-IDF Vectorizer]
        Cosine[Cosine Similarity]
        Heuristics[Structural Heuristics]
    end
    
    subgraph "Validation"
        Benchmarks[(Supabase Benchmarks)]
        Team_DB[(Team History)]
    end
    
    PR_Text --> TF_IDF
    TF_IDF --> Cosine
    Benchmarks --> Cosine
    Team_DB --> Cosine
    
    PR_Meta --> Heuristics
    
    Cosine --> Final[Weighted Risk Score]
    Heuristics --> Final
    
    Final --> UI_Output[Interactive Panel]

Loading

---

🌊 Logic Flow

When you open a PR, PitStop AI springs into action:

sequenceDiagram
    participant User
    participant Bitbucket
    participant Forge
    participant ML as ML Service
    participant Gemini as Gemini AI
    
    User->>Bitbucket: View Pull Request
    Bitbucket->>Forge: Load Risk Panel
    Forge->>Forge: Fetch PR Diff & Metadata
    
    par ML Analysis
        Forge->>ML: Vectorize PR Data
        ML->>ML: Compare vs Benchmark PRs
        ML-->>Forge: Statistical Risk Score
    and AI Analysis
        Forge->>Gemini: Analyze PR Logic
        Gemini-->>Forge: Remediation Suggestions
    end
    
    Forge-->>Bitbucket: Populate Panel
    Bitbucket-->>User: Show Analysis & Recommendations

Loading

---

🚀 Getting Started

1. Prerequisites

• Forge CLI installed.
• Bitbucket Cloud workspace.
• Supabase project (for benchmarking data).

2. Installation

1. Clone the Repo

   git clone https://github.com/Shafwansafi06/CodeRiskRadar.git
   cd CodeRiskRadar

2. Setup Frontend

   cd frontend
   npm install && npm run build
   cd ..

3. Deploy to Forge

   forge deploy
   forge install      # Select your Bitbucket site

4. Variables Configuration

   forge variables set SUPABASE_URL your_url
   forge variables set SUPABASE_SERVICE_KEY your_key
   forge variables set GEMINI_API_KEY your_key

---

🔐 Security & Privacy

PitStop AI is designed for enterprise-grade security:

• No Code Leakage: Your code diffs are processed in the volatile memory of the Forge runtime.
• Atlassian Egress: All outgoing connections are explicitly declared in the manifest.yml.
• Telemetry: Uses the ObservAI SDK for secure LLM monitoring without content logging.

---

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

---

MIT © 2025 — PitStop AI | Created by Shafwan Safi 🚀