def run_rag_chain(question_text, retriever_obj, prompt_template):
...
rag_chain = (
RunnableParallel( clean_context,
"query": RunnablePassthrough()
)
| immediate
| llm
| StrOutputParser()
)
return rag_chain.invoke(question_text)
The RAG chain handles every thing: fetching the correct context, structuring the immediate, producing a response, and parsing the output.
def calculate_confidence_score(reply: str, context_chunks: record):
"""Calculate confidence rating for solutions"""
embedder = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2')
answer_embedding = embedder.embed_query(reply)
context_text = " ".be part of([chunk.page_content for chunk in context_chunks])
context_embedding = embedder.embed_query(context_text)rating = cosine_similarity(
[np.array(answer_embedding)],
[np.array(context_embedding)]
)[0][0]
return spherical(float(rating), 2)
Rationalization:
This perform checks how semantically related the generated reply is to the unique doc chunks utilizing cosine similarity. It returns a rating between 0 and 1 with labels: Low / Medium / Excessive.
ScholarMate defines every tutorial assistant characteristic as a LangChain Software. These instruments summary away the retrieval and technology steps utilizing @instrument decorators and depend on run_rag_chain() internally.
