Metadata-Version: 2.4
Name: astracompass
Version: 0.1.0
Summary: A modular framework for evaluating, training, and improving large language models.
Author: Irena Popova
License-Expression: MIT
Project-URL: Homepage, https://github.com/irenapopova/astra
Project-URL: Repository, https://github.com/irenapopova/astra
Project-URL: Issues, https://github.com/irenapopova/astra/issues
Keywords: llm,large-language-models,evaluation,model-training,fine-tuning,semantic-analysis,model-adaptation,machine-learning,artificial-intelligence
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3.14
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Dynamic: license-file

# Astra
Astra is a modular framework for evaluating, fine-tuning, and improving large language models (LLMs) for domain-specific knowledge and real-world business tasks.
# Astra

### An LLM Evaluation, Training & Adaptation Framework

Astra is an open-source project designed to support the evaluation, training, fine-tuning, and continuous improvement of large language models (LLMs).

The project focuses on identifying model weaknesses, improving domain-specific understanding, and developing more reliable AI systems capable of performing complex, specialized tasks.

Rather than relying exclusively on general-purpose model capabilities, Astra aims to provide a structured approach to understanding how language models perform, where they fail, and which techniques can improve their performance.

## Core Objectives

**Model Evaluation**

Evaluate model performance across domain-specific tasks, including factual accuracy, reasoning, semantic understanding, instruction following, and response consistency.

**Error Detection & Analysis**

Identify recurring mistakes, knowledge gaps, ambiguous interpretations, unsupported claims, and weaknesses in task execution.

**Domain-Specific Adaptation**

Develop datasets, evaluation criteria, and adaptation workflows for specialized industries and business functions.

**Model Training & Fine-Tuning**

Support supervised fine-tuning and other training techniques using curated datasets and measurable performance objectives.

**Continuous Improvement**

Compare model versions, measure improvements against established benchmarks, and identify regressions after training or configuration changes.

## Planned Capabilities

- Configurable evaluation pipelines and benchmark suites
- Structured error classification and reporting
- Domain-specific dataset creation and validation
- Semantic and contextual reasoning assessments
- Integration with open-source and API-based language models
- Training and fine-tuning workflows for supported models
- Model comparison and performance tracking
- Experiment management and reproducible results

## Applications

Astra is intended to support AI applications across business operations, education, professional services, knowledge management, customer communication, and other specialized domains.

## Project Status

**Early development  architecture and research phase.**

The initial focus is on building the evaluation infrastructure, developing a systematic error taxonomy, and establishing measurable criteria for model improvement. Training and adaptation capabilities will be developed as subsequent components.

## Vision

To make language model development more systematic, transparent, measurable, and adaptable to the knowledge and requirements of different domains.
