shaaha 2.0.1

Intelligent Meta-Dispatcher — AI-powered backend selection, agent execution, and code optimization.

🧠 Shaaha

One import. Best backend. Always. Now with AI.

Shaaha is the world's first Self-Optimizing AI-Driven Python Meta-Dispatcher.

Instead of manually picking numpy, pandas, torch, or sklearn — you write one import and Shaaha silently picks the best available library for your hardware, learns from every run, fixes its own errors, and even rewrites your old code to be faster.

---

🚀 Install

pip install shaaha                      # zero dependencies
pip install "shaaha[default]"           # + numpy, pandas, pillow, sklearn, matplotlib
pip install "shaaha[science]"           # full scientific stack
pip install "shaaha[gpu]"               # CUDA / GPU stack

---

⚡ Quick Start

import shaaha

import shaaha.math as sm           # → cupy (GPU) or numpy (CPU)
import shaaha.dataframe as df      # → polars (large) or pandas (small)
import shaaha.ml as ml             # → torch (GPU) or sklearn (CPU)
import shaaha.image as img         # → opencv or pillow
import shaaha.viz as viz           # → plotly or matplotlib

arr = sm.array([1, 2, 3, 4, 5])
sm.sqrt(arr)

model = ml.LinearRegression()
model.fit(X_train, y_train)

---

🤖 Layer 2 — AI Agent

shaaha.agent("load sales.csv, find the top 10 products, and plot a bar chart")
# Breaks task into steps and executes automatically
# 📄 Report saved: shaaha_report.md

---

🔁 Layer 3 — Code Rewriter

shaaha.optimize_file("my_old_script.py")
# Finds slow patterns, shows diff, asks before applying
# ⚡ Estimated speedup: up to 100x faster

---

🧠 Layer 4 — Adaptive Brain

shaaha.status()
# "brain_confidence": "87%"
# "recommendations": ["math: 'numpy' fastest on your machine (4.2ms)"]

shaaha.reset_learning()

---

🩺 Layer 5 — Self-Healer

When a backend crashes, Shaaha auto-switches and always notifies you:

⚠️  'jax' failed → switched to 'numpy'.
   shaaha.suppress_warnings = True to hide this.

---

📖 Layer 6 — Explainer

shaaha.explain("my_script.py", output="report.md")
# Explains every library choice in plain English
# Perfect for students and researchers

---

📊 Layer 7 — Dashboard

shaaha.dashboard()
# 🚀 Opens http://127.0.0.1:7842
# Shows all operations, backends chosen, speed history

---

🔌 Layer 8 — Plugin System

shaaha.register_backend("dataframe", "vaex", "vaex", priority=88)
shaaha.list_backends("dataframe")

---

🤝 Layer 9 — Collaboration

shaaha.share_profile("my_profile.json")
shaaha.import_profile("teammate.json")
shaaha.sync("http://team-server:8080")

---

🛡️ Layer 10 — Safety Guard

shaaha.safe_mode(True)
# ✅ Results match 100%. Safe to switch from 'jax' to 'numpy'.

---

🗺️ Supported Domains

shaaha.<domain>	Best Backend	Fallbacks
shaaha.math	cupy / jax	torch → numpy
shaaha.dataframe	polars	modin → pandas → dask
shaaha.ml	torch	xgboost → lightgbm → sklearn
shaaha.image	opencv	pillow → skimage
shaaha.nlp	transformers	spacy → nltk
shaaha.viz	plotly	altair → seaborn → matplotlib
shaaha.stats	statsmodels	scipy → sklearn
shaaha.http	httpx	requests → urllib3
shaaha.json	orjson	ujson → json (stdlib)

---

🏗️ How It Works

Shaaha uses PEP 302 Import Hooks (sys.meta_path) to intercept Python's import system before it looks for any file. A MetaPathFinder captures every import shaaha.* call, the Router scores available backends by hardware context and learned timing data, and a lazy ProxyModule forwards all attribute access transparently — zero cost until first use.

---

🎓 MSc Thesis

Shaaha implements three advanced CS concepts:

1. Python Import Protocol (PEP 302/451) — MetaPathFinder, Loader, ModuleSpec
2. Proxy + Strategy + Registry Design Patterns
3. Context-Aware Weighted Dispatch with Adaptive Learning

---

📄 License

MIT © Shaaha | pypi.org/project/shaaha