davex-ai/Vector-Alpha-Model

🚀 ARCHON

---

🧠 What is this?

A system that analyzes any GitHub repository and generates deep technical interview questions + answers based on:

• Architecture
• Scalability
• Tradeoffs
• Real-world engineering decisions

⚡ Works even without LLM access using a fallback heuristic engine.

---

✨ Features

🔍 Repository Analysis

• Fetches and parses GitHub repositories via API
• Prioritizes important files (core logic > boilerplate)
• Supports multiple languages

🧩 Intelligent Chunking

• Breaks code into meaningful chunks
• Filters noise (non-informative code)
• Preserves structural context

🧠 Embedding + Retrieval (RAG)

• Uses SentenceTransformers
• Retrieves most relevant code sections
• Builds contextual understanding of system design

🤖 AI Question Generation

• Generates interview-level questions on:

  • Architecture decisions
  • Scalability concerns
  • Tradeoffs

⚡ Fallback Mode (No LLM Required)

• Automatically switches to rule-based generation

• Uses detected signals:

  • API usage
  • State management
  • Auth systems
  • Async logic

---

🧱 System Architecture

GitHub Repo
   ↓
File Fetching + Prioritization
   ↓
Chunking + Filtering
   ↓
Embeddings (SentenceTransformers)
   ↓
Vector Similarity Retrieval
   ↓
Context Builder
   ↓
AI Question Generator
   ↓
Fallback Engine (if LLM unavailable)

---

🛠️ Tech Stack

Backend:
- FastAPI
- Python

AI / ML:
- SentenceTransformers
- Cosine Similarity (Sklearn)

Data:
- GitHub REST API

Frontend:
- Minimal Web UI (React / HTML)

Optional:
- OpenAI API (LLM generation)

---

⚙️ How it Works

1. Input a GitHub repo URL

2. System fetches and filters key files

3. Code is chunked and embedded

4. Relevant chunks are retrieved

5. Questions are generated using:

   • LLM (if available)
   • OR fallback heuristic engine

---

📡 API Usage

POST /analyze

{
  "repo_url": "https://github.com/user/repo",
  "num_questions": 5
}

Response

{
  "repo": "...",
  "mode": "mock",
  "questions": [
    {
      "id": 1,
      "question": "...",
      "answer": "..."
    }
  ]
}

---

⚠️ Challenges Solved

• Large repo handling (chunking + prioritization)
• Token limitations (retrieval instead of full context)
• LLM dependency → solved with fallback system
• Noise reduction in code analysis

---

💡 Future Improvements

• 🔥 Dynamic repo-type detection (ML, backend, real-time, etc.)
• 📊 Question difficulty levels (junior → senior)
• 🔗 Follow-up interview questions
• 🧠 Hybrid LLM + rule-based reasoning
• ⚡ Caching + performance optimization

---

🧑‍💻 Author

Built by Dave — aspiring systems engineer ⚡

---

🎬 Demo

---

⭐ Support

If this project helped or inspired you:

• ⭐ Star the repo
• 🍴 Fork it
• 🧠 Build something even crazier

---

“Don’t just read code. Interrogate it.”