Akairo

Akairo

DSPy-based optimizer for LLM responses. Supports local models via Ollama and cloud models via OpenAI.

Architecture

Path	Purpose
src/dspy_local_optimizer/core/models.py	Data models
src/dspy_local_optimizer/core/metrics.py	Quality scoring
src/dspy_local_optimizer/core/ollama.py	Ollama client
src/dspy_local_optimizer/core/openai.py	OpenAI client
src/dspy_local_optimizer/optimizers/guideline_optimizer.py	Guideline optimization logic
scripts/run_optimization.py	CLI entry point
tests/test_copro_optimizer.py	1 test file

Requirements

• Python 3.10+ with Poetry
• Ollama (for local models) or an OpenAI API key

Setup

git clone https://github.com/jmanhype/Akairo.git
cd Akairo
poetry install

# For local models:
ollama pull llama2

Usage

from dspy_local_optimizer import BatchOptimizedGuidelineManager

optimizer = BatchOptimizedGuidelineManager(
    model_name="ollama/llama2",  # or "openai/gpt-3.5-turbo"
    use_optimizer=True
)

optimized = optimizer.optimize_guidelines(
    guidelines=your_guidelines,
    examples=training_data,
    batch_size=5
)

Or via the script:

poetry run python scripts/run_optimization.py

Tests

poetry run pytest

1 test file.

Status

Small utility project. The optimization loop runs but there are no published benchmarks showing improvement over baseline. The codebase is ~6 source files.

License

MIT