Lesson 08: Service Layer#

"Business logic belongs in services, not routers."

Introduction#

The service layer contains all business logic - OpenAI calls, workflows, document processing, and monitoring.

Services Overview#

1. openai_service.py - LLM interactions
2. monitoring_service.py - Cost tracking
3. multi_agent_email_service.py - Workflow orchestration
4. document_service.py - RAG operations
5. rag_graph_service.py - LangGraph workflows

openai_service.py Functions#

generate_agent_response()#

Purpose: Chat completion with OpenAI Parameters: system_prompt, user_message, agent_name, language, tone Returns: str (assistant response)

analyze_email()#

Purpose: Extract structured data from email Returns: dict with summary, intent, priority, deadline, action_items, suggested_reply

generate_rag_answer()#

Purpose: Generate answer from retrieved context Parameters: question, contexts, agent_name, language, tone

evaluate_retrieved_context()#

Purpose: Check if context contains answer Returns: dict with has_answer, confidence, reason

monitoring_service.py#

estimate_tokens()#

Approximates token count (1 token ≈ 4 characters)

estimate_openai_cost()#

Calculates cost based on model pricing

multi_agent_email_service.py#

run_analysis_agent()#

Extract structured data from email

run_reply_agent()#

Generate professional reply

run_reviewer_agent()#

Check quality and approve/reject

document_service.py#

split_text_into_chunks()#

Breaks documents into segments (800 chars with 100 overlap)

create_embedding()#

Generates OpenAI embedding vector

add_chunk_to_vector_store()#

Saves to ChromaDB

search_agent_documents()#

Semantic search for relevant chunks

rag_graph_service.py#

RAGGraphState#

TypedDict defining workflow state

Nodes#

• retrieve_node: Fetch chunks
• evaluate_node: Check quality
• answer_node: Generate response
• refusal_node: Return "don't know"

build_corrective_rag_graph()#

Constructs LangGraph state machine

Key Takeaways#

🎯 Services contain business logic separated from HTTP 🎯 Each service wraps specific functionality 🎯 Services are testable without HTTP clients 🎯 LangGraph enables complex workflows

Next Steps#

👉 Lesson 09: API Routers →

---

Module 2 · Data Layer