pyproject.toml

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[project]
name = "llmops-research-assistant"
version = "0.1.0"
description = "Production-architecture LLMOps platform: RAG, fine-tuning, evaluation, and deployment"
readme = "README.md"
license = { file = "LICENSE" }
requires-python = ">=3.11"
authors = [{ name = "Joseph Ahn", email = "josephahn63@gmail.com" }]
keywords = ["llm", "rag", "mlops", "fine-tuning", "inference", "langchain", "langgraph"]
classifiers = [
    "Development Status :: 4 - Beta",
    "Intended Audience :: Developers",
    "Topic :: Scientific/Engineering :: Artificial Intelligence",
    "Programming Language :: Python :: 3.11",
    "Programming Language :: Python :: 3.12",
]

# ── Core dependencies ──────────────────────────────────────────────────────────
dependencies = [
    # API
    "fastapi>=0.111.0",
    "uvicorn[standard]>=0.29.0",
    "websockets>=12.0",
    "pydantic>=2.7.0",

    # LLM + embeddings
    "openai>=1.30.0",
    "tiktoken>=0.7.0",
    "sentence-transformers>=3.0.0",
    "transformers>=4.41.0",
    "tokenizers>=0.19.0",

    # RAG + retrieval
    "faiss-cpu>=1.8.0",
    "rank-bm25>=0.2.2",
    "langchain>=0.2.0",
    "langchain-core>=0.2.0",
    "langchain-openai>=0.1.0",
    "langgraph>=0.1.0",
    # ragas<=0.4.3 imports langchain_community.chat_models.vertexai at module load
    # time; that submodule was removed in langchain-community 0.4.2. Pin below that
    # until ragas drops the import (unreleased upstream fix as of this writing).
    "langchain-community>=0.2.0,<0.4.2",

    # Evaluation
    "ragas>=0.1.0",
    "deepeval>=0.21.0",

    # MLflow + experiment tracking
    "mlflow>=2.13.0",

    # Async task queue
    "celery[redis]>=5.4.0",
    "redis>=5.0.0",

    # Data
    "numpy>=1.26.0",
    "pandas>=2.2.0",
    "umap-learn>=0.5.0",

    # Cloud
    "boto3>=1.34.0",

    # Observability
    "prometheus-client>=0.20.0",

    # Config
    "hydra-core>=1.3.0",
    "omegaconf>=2.3.0",
    "pyyaml>=6.0.0",

    # Infra
    "docker>=7.0.0",
]

# ── Optional dependency groups ─────────────────────────────────────────────────
[project.optional-dependencies]

# GPU inference
gpu = [
    "torch>=2.3.0",
    "torchvision>=0.18.0",
    "vllm>=0.4.0",
    "accelerate>=0.30.0",
    "bitsandbytes>=0.43.0",
    "peft>=0.11.0",
]

# Local inference (llama.cpp)
local = [
    "llama-cpp-python>=0.2.70",
    "gguf>=0.8.0",
]

# Spark / Databricks
spark = [
    "pyspark>=3.5.0",
    "delta-spark>=3.1.0",
]

# Streaming
streaming = [
    "kafka-python>=2.0.2",
    "aiokafka>=0.11.0",
]

# Azure
azure = [
    "azure-search-documents>=11.6.0",
    "azure-core>=1.30.0",
    "azure-identity>=1.16.0",
]

# Evaluation UI
ui = [
    "gradio>=4.36.0",
    "websockets>=12.0",
]

# ONNX / inference optimization
inference = [
    "onnx>=1.16.0",
    "onnxruntime-gpu>=1.18.0",
]

# Multimodal
multimodal = [
    "diffusers>=0.27.2",
    "Pillow>=10.3.0",
]

# ML classifiers (adversarial, behavioral, interpretability probes)
ml = [
    "scikit-learn>=1.5.0",
    "sentence-transformers>=3.0.0",
]

# RL environments
rl = [
    "gymnasium>=0.29.0",
    "stable-baselines3>=2.3.0",
]

# Tokenization extras
tokenization = [
    "sentencepiece>=0.2.0",
    "tokenizers>=0.19.0",
]

# Recommendation system (learn-to-rank, SHAP)
recsys = [
    "lightgbm>=4.3.0",
    "shap>=0.45.0",
    "scikit-learn>=1.5.0",
]

# Causal inference
causal = [
    "numpy>=1.26.0",
]

# Cross-model validation (OpenAI + Anthropic truth committee)
truth_committee = [
    "anthropic>=0.34.0",
]

# Dev tools
dev = [
    "pytest>=8.2.0",
    "pytest-asyncio>=0.23.0",
    "pytest-cov>=5.0.0",
    "ruff>=0.4.0",
    "mypy>=1.10.0",
    "pre-commit>=3.7.0",
    "httpx>=0.27.0",
]

# All optional deps (for full dev environment)
all = [
    "llmops-research-assistant[gpu,local,spark,streaming,azure,ui,inference,multimodal,ml,rl,tokenization,recsys,causal,truth_committee,dev]",
]

# ── CLI entry points ───────────────────────────────────────────────────────────
[project.scripts]
llmops-serve    = "api.main:serve"
llmops-eval     = "mlops.ragas_tracker:main"
llmops-ingest   = "ingestion.pipeline:main"
llmops-ui       = "eval.gradio_eval_ui:main"
llmops-research-ui = "eval.research_ui:main"
llmops-config   = "config.hydra_config:main"

# ── Tool configuration ─────────────────────────────────────────────────────────

[tool.hatch.build.targets.wheel]
packages = [
    "agents", "analysis", "api", "causal", "cicd", "compile", "config", "context_engineering",
    "prompt_registry",
    "cuda_ext", "eval", "experiments", "finetune", "governance", "inference", "ingestion", "interpretability",
    "mcp_server", "microservices", "mlops", "monitoring",
    "multimodal", "observability", "recsys", "rl", "safety",
    "sandbox", "spark_ml", "streaming", "tokenization",
]

[tool.ruff]
line-length = 100
target-version = "py311"
select = ["E", "F", "I", "N", "W", "UP", "B", "C4", "PT"]
ignore = ["E501", "B008"]

[tool.ruff.per-file-ignores]
"tests/*" = ["S101"]

[tool.mypy]
python_version = "3.11"
ignore_missing_imports = true
strict = false
warn_unused_ignores = true

[tool.pytest.ini_options]
asyncio_mode = "auto"
testpaths = ["tests"]
addopts = "--cov=. --cov-report=term-missing --cov-report=xml -v"
filterwarnings = ["ignore::DeprecationWarning"]

[tool.coverage.run]
omit = ["tests/*", "notebooks/*", "benchmarks/*", "conf/*", "outputs/*"]

[tool.uv]
dev-dependencies = [
    "ruff>=0.4.0",
    "mypy>=1.10.0",
    "pytest>=8.2.0",
    "pytest-asyncio>=0.23.0",
    "pytest-cov>=5.0.0",
    "pre-commit>=3.7.0",
    "httpx>=0.27.0",
]