🤖 AI-Native Selenium Framework Assistant

A Retrieval-Augmented AI system designed for architecture-aware automation engineering.

---

✨ Why This Project Exists

Modern LLMs are powerful — but in automation engineering, generic AI responses can:

• Invent framework methods
• Assume incorrect driver setup patterns
• Break Page Object Model (POM) structure
• Generate non-aligned test code

Automation frameworks demand architectural precision — not hallucinated code. This project solves that problem.

---

🧠 What This Assistant Does

Instead of answering blindly, this system:

1. Scans your automation framework source code.
2. Splits it into method-level chunks.
3. Creates semantic embeddings locally.
4. Stores them in a persistent vector database (ChromaDB).
5. Retrieves relevant code before querying the LLM.
6. Forces the LLM to answer strictly using retrieved context.

Note: If something is not implemented, it responds: "Not implemented in this framework." No guessing. No fabrication.

---

🏗 System Architecture

• Automation Framework Source Code
• ↓ Method-Level Chunking
• ↓ SentenceTransformer Embeddings
• ↓ ChromaDB (Persistent Vector Store)
• ↓ Top-K Semantic Retrieval
• ↓ Groq LLM (LLaMA 3.1)
• ↓ Grounded, Framework-Aware Response

---

🎯 Example Queries

• Where is the WebDriver instance created?
• Is retry logic implemented?
• Explain the login page implementation.
• Generate a test aligned with the POM structure.
• Validate architectural consistency.

---

⚙️ Quick Start

1️⃣ Install Dependencies

pip install -r requirements.txt

2️⃣ Configure Environment

Create a .env file in the root directory:

GROQ_API_KEY=your_api_key_here
FRAMEWORK_PATH=path_to_your_automation_framework

3️⃣ Generate Embeddings

python create_local_vector.py

4️⃣ Run the Assistant

python ai_assistant.py

---

🔐 Security & Clean Design

• API keys are not committed.
• vector_db/ is ignored via .gitignore.
• Framework source code is not included in the repo.
• Configurable framework path via .env.
• Strict grounding enforced.

---

🧩 Design Philosophy

• Strict grounding (no hallucination).
• Architecture-aware AI responses.
• Clean separation of concerns.
• Controlled token usage.
• Scalable RAG design principles.
• Reliability over hype.

---

📊 Current Status

Phase 1 — Stable Grounded RAG Assistant (Implemented)

• Method-level chunking
• Metadata-aware embeddings
• Persistent vector storage
• Top-K semantic retrieval
• Strict LLM response constraints
• Clean CLI interface

Phase 2 — Future Roadmap (Planned)

• Stacktrace failure analyzer
• CI pipeline integration
• Hybrid retrieval (vector + keyword)
• Architecture review mode

---

🚀 Vision

Move toward AI-Native Automation Engineering, where LLM systems assist in debugging, onboarding, and maintaining architectural consistency without compromising structure or reliability.

---

🏷 Tech Stack

• Language: Python
• Embeddings: SentenceTransformers
• Vector DB: ChromaDB
• LLM: Groq (LLaMA 3.1)
• Domain: Selenium Automation Framework