Metadata-Version: 2.4
Name: krysta-wing
Version: 1.0.2
Summary: A unified multimodal model evaluation tracking and engineering report engine.
Author: Anshu Aditya
Project-URL: Homepage, https://github.com/your-username/kwing_library
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Requires-Python: >=3.8
Description-Content-Type: text/markdown
Requires-Dist: jinja2>=3.0.0
Requires-Dist: matplotlib>=3.5.0
Requires-Dist: numpy>=1.20.0
Requires-Dist: pyyaml>=6.0

# Krysta Wing (`krysta-wing`)

A minimalist, local telemetry and reporting engine designed to audit machine learning models, log multi-modal interpretability artifacts, and flag performance regressions locally.

---

## Core Architecture

Krysta Wing provides a lightweight alternative to heavyweight, cloud-dependent MLOps platforms. It executes entirely in your local runtime environment, tracking performance metrics across iterations and automatically identifying anomalies using moving statistical baselines ($\mu \pm 2\sigma$).

### Key Capabilities
* **Model-Agnostic Routing:** Native abstraction paths for handling spatial feature maps, attention heatmaps (e.g., Grad-CAM), and NLP token confidence streams.
* **Statistical Regression Detection:** Automated tracking engine that runs calculations against historical baselines to intercept memory leaks or latency spikes.
* **Configuration-Driven Portability:** Decoupled architecture managing environment settings, directory structures, and alerting boundaries via standard YAML files.

---

## Installation

Install the package directly from PyPI:

```bash
pip install krysta-wing

Quick Start
1. Define Environment Constraints
Create a kwing_config.yaml file in your root working directory to set evaluation thresholds and report destination directories dynamically:
```python
workspace_root: "production_reports"

thresholds:
  token_confidence: 0.85
  latency_limit_ms: 50.0
```

2. Instrumentation Pipeline
Integrate the telemetry tracking wrapper into your evaluation loop:
```python

import numpy as np
from kwing_reporter import ModelReport

reporter = ModelReport(week=22, model_name="ResNet50-XAI", modality="hybrid-omni")

# Log performance telemetry
reporter.metrics = {
    "latency": 14.2,   # in milliseconds
    "vram": 3120.0,    # in Megabytes
    "loss": 0.042
}


tokens = ["Initiating", "attention", "map"]
confidences = [0.94, 0.72, 0.91]

reporter.log_custom_artifact(
    data=tokens,
    artifact_type="tokens",
    title="Layer 4 Token Confidence Pass",
    confidences=confidences,
    sample_phrase=" ".join(tokens)
)


reporter.compile()
```


License
Distributed under the MIT License. See LICENSE for details.
