from haystack import Pipeline
from haystack.components.generators import OpenAIGenerator
from haystack.components.builders import PromptBuilder
from haystack.components.retrievers.in_memory import InMemoryBM25Retriever
from haystack.document_stores.in_memory import InMemoryDocumentStore

# Setup document store
doc_store = InMemoryDocumentStore()
doc_store.write_documents(documents)

# Initial generic prompt
initial_prompt = """
Context: {{context}}
Question: {{question}}
Answer:
"""

# Build pipeline
pipeline = Pipeline()
pipeline.add_component("retriever", InMemoryBM25Retriever(document_store=doc_store))
pipeline.add_component("prompt_builder", PromptBuilder(template=initial_prompt))
pipeline.add_component("generator", OpenAIGenerator(model="gpt-4o-mini"))

pipeline.connect("retriever", "prompt_builder.context")
pipeline.connect("prompt_builder", "generator")

import dspy

class HaystackRAG(dspy.Module):
    """DSPy module wrapping Haystack retriever."""
    
    def __init__(self, retriever, k=3):
        super().__init__()
        self.retriever = retriever
        self.k = k
        self.generate = dspy.ChainOfThought("context, question -> answer")
    
    def forward(self, question):
        # Use Haystack retriever
        results = self.retriever.run(query=question, top_k=self.k)
        context = [doc.content for doc in results['documents']]
        
        # Use DSPy for generation
        pred = self.generate(context=context, question=question)
        return dspy.Prediction(context=context, answer=pred.answer)

from haystack.components.evaluators import SASEvaluator

# Haystack semantic evaluator
sas_evaluator = SASEvaluator(model="sentence-transformers/all-MiniLM-L6-v2")

def mixed_metric(example, pred, trace=None):
    """Combine semantic accuracy with conciseness."""
    
    # Semantic similarity (Haystack SAS)
    sas_result = sas_evaluator.run(
        ground_truth_answers=[example.answer],
        predicted_answers=[pred.answer]
    )
    semantic_score = sas_result['score']
    
    # Conciseness penalty
    word_count = len(pred.answer.split())
    conciseness = 1.0 if word_count <= 20 else max(0, 1 - (word_count - 20) / 50)
    
    return 0.7 * semantic_score + 0.3 * conciseness

from dspy.teleprompt import BootstrapFewShot

lm = dspy.LM("openai/gpt-4o-mini")
dspy.configure(lm=lm)

# Create DSPy module with Haystack retriever
rag_module = HaystackRAG(retriever=pipeline.get_component("retriever"))

# Optimize
optimizer = BootstrapFewShot(
    metric=mixed_metric,
    max_bootstrapped_demos=4,
    max_labeled_demos=4
)

compiled = optimizer.compile(rag_module, trainset=trainset)