pynlqe 0.1.0

Natural language to SQL query engine powered by LangChain and DuckDB

NLQE (Natural Language NLQE)

A natural language to SQL query engine powered by LangChain and DuckDB. Ask questions about your data in plain English — works with OpenAI, Anthropic, or any LangChain-compatible model.

Version: v0.1.0 | Python: 3.11+ | License: MIT

---

Examples

Single query

from nlqe import QueryEngine, QueryEngineConfig

# Reads NLQE_OPENAI_API_KEY from environment or .env
config = QueryEngineConfig()
engine = QueryEngine(config)
engine.load_datasource("transactions.parquet")

response = engine.query("What was total revenue by region last month?")
print(response.answer)
# "North America led with $1.2M, followed by Europe at $890K ..."

print(f"SQL: {response.generated_sql}")
print(f"Rows: {len(response.data)}  Confidence: {response.confidence_score:.0%}")

Multi-turn conversation

conv = engine.start_conversation()

r1 = conv.query("Show me the top 5 products by revenue")
r2 = conv.query("Which of those had the highest return rate?")  # uses context from r1
r3 = conv.query("Compare that to last month")

print(r3.answer)

Switch to LLM Providers

from langchain_anthropic import ChatAnthropic
from nlqe import QueryEngine, QueryEngineConfig
from nlqe.llm import LLMClient

# Anthropic
llm = ChatAnthropic(model="claude-3-5-sonnet-20241022")
engine = QueryEngine(QueryEngineConfig(), custom_llm_client=LLMClient(llm))

# Ollama (Local)
from langchain_ollama import ChatOllama
engine = QueryEngine(QueryEngineConfig(), custom_llm_client=LLMClient(ChatOllama(model="llama3")))

---

Installation

# Install core package
pip install pynlqe

# Install with development tools
pip install "pynlqe[dev]"

Generate the sample dataset for testing:

python create_sample_data.py

Configuration

Copy .env.example to .env and fill in your credentials:

cp .env.example .env

NLQE_LLM_PROVIDER=openai
NLQE_OPENAI_API_KEY=sk-...
NLQE_LLM_MODEL=gpt-4o

---

Overview

NLQE translates plain English into SQL executed against structured data (Parquet, CSV) via DuckDB. It uses an iterative debug loop to automatically recover from SQL errors.

Key features:

• Swappable LLM providers (OpenAI, Anthropic, Ollama, etc.)
• Multi-turn conversations with context preservation
• Automatic SQL error recovery (Iterative Debug Loop)
• Robust evaluation framework with "golden datasets"
• High test coverage and strict type safety

---

Architecture Overview

Component	Location	Purpose
QueryEngine	engine.py	Main entry point
LLMClient	llm/client.py	SQL generation, debugging, and synthesis
DuckDBExecutor	duckdb/executor.py	Safe SQL execution via DuckDB
QueryLoop	query/loop.py	Orchestration logic
ConversationManager	conversation/manager.py	Multi-turn history tracking

For more details see ARCHITECTURE.md.

---

Development

Use the provided Makefile for common development tasks:

make install    # Install dependencies
make lint       # Run ruff and mypy
make format     # Format code with ruff
make test       # Run all tests
make build      # Build distribution packages

Running the Evaluation Suite

Evaluate the engine against standardized test cases:

python -m nlqe.testing.cli evaluate --dataset fixtures/golden_datasets.yaml

---

Documentation

Document	Description
API.md	Public API reference and detailed usage
ARCHITECTURE.md	Technical design and component data flows
DESIGN.md	High-level goals and philosophy
TESTING.md	Evaluation strategy and accuracy metrics
FAQ.md	Design decisions and common questions

---

Future Roadmap

• v0.2.0: Support for PostgreSQL and Snowflake datasources, result caching, and custom synthesizers.
• v1.0.0: Cloud-native API, fine-tuning support, and advanced production monitoring.

---

License

This project is licensed under the MIT License.