=============================================
    ___ _    ___________           ____   ___
   /   | |  / /  _/ ___/   _   __ / __ \ <  /
  / /| | | / // / \__ \   | | / // / / / / / 
 / ___ | |/ // / ___/ /   | |/ // /_/ / / /  
/_/  |_|___/___//____/    |___(_)____(_)_/   
                                             
=============================================

🚀 Avis: Agentic Code Generation System

Avis is an LLM-based code generation agent purpose-built to deliver accurate, production-ready code using a Retrieval-Augmented Generation (RAG) pipeline over framework documentation.

📌 Mission: To eliminate hallucination and boost developer productivity by grounding every response in real, version-aware documentation.

---

📦 What It Does

Avis is designed to:

• Ingest and embed framework documentation (e.g., LangGraph, FastAPI, PyTorch).
• Retrieve highly relevant chunks using hybrid retrieval and custom rerankers.
• Generate precise code using instruction-following LLMs (OpenAI, Gemini, Claude, etc.).
• Automatically self-evaluate, rerun, and correct outputs to ensure consistent reliability.

---

🧠 System Overview

User Prompt
│
▼
[Input Parser] → [Retriever (Semantic + Filtered)] → [Reranker]
│ │
└────────────────────→ [Code Generation Agent] ←───┘
│
▼
[Self-Checker + Rewriter Loop]
│
▼
Final Code Output ✅

---

⚙️ Key Features

• 📚 Framework-Aware RAG:
  Parses and chunks docs using custom strategies, retaining structure (functions, classes, examples).

• 🔎 Hybrid Retrieval:
  FAISS + keyword filters + metadata (e.g., version, module, type).

• 🧠 Agentic Code Loop:

  • observe: Analyze prompt & retrieve context
  • think: Choose generation strategy (e.g., template, scratch, edit)
  • act: Generate code
  • evaluate: Check syntax/errors/tests
  • revise: Rerun or correct if needed

• 📈 Accuracy-First Design:

  • Zero-shot eval
  • AST comparisons
  • Task-level metrics (pass@1, correctness, hallucination rate)

• 🔌 Plug & Play Support for LLMs:
  OpenAI GPT-4, Claude Opus, Gemini 1.5, Mistral, etc.

---

🔧 Setup

# Clone the repo
git clone https://github.com/your-org/avis.git && cd avis

Setup environment

python -m venv .venv && source .venv/bin/activate pip install -r requirements.txt

# (Optional) Install FAISS
pip install faiss-cpu

#📁 Project Structure

avis/
├── indexer/            # Chunking + embedding from docs
├── retriever/          # Vector + filter-based retrieval
├── agent/              # LLM orchestration + evaluation loop
├── server/             # FastAPI wrapper for web-based usage
├── examples/           # Prompt → Code → Eval demos
├── configs/            # YAML configs for pipelines
└── tests/              # Eval + test cases

#🚀 Usage

1. Embed Documentation

python -m avis.indexer --config configs/langgraph.yml --docs ./docs/langgraph/

2. Run Server

uvicorn server.main:app --reload

3. Query Agent (via API or CLI)

curl -X POST http://localhost:8000/code \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Write a LangGraph node that retries if error occurs"}'

✅ Sample Output Prompt:

Create a LangGraph node that retries a step on error with exponential backoff.

Output:

from langgraph.graph import StateGraph
from tenacity import retry, wait_exponential, stop_after_attempt

@retry(wait=wait_exponential(min=1, max=10), stop=stop_after_attempt(3))
def risky_node(state):
    # Your logic here
    return state

builder = StateGraph()
builder.add_node("retry_node", risky_node)

✅ Backed by docs: langgraph/StateGraph, tenacity.retry, etc. ✅ Passed syntax + lint check

📌 Roadmap GitHub Copilot integration

Multi-version doc support

Auto-patch hallucinated code

VSCode extension

📄 License MIT License. See LICENSE for details.

Made with ⚡ by [S.N.K] — AI for Devs, Done Right.