Module 3: Services Deep Dive#

Overview#

Module 2 gave you the map; this module is the territory. Every one of the five services gets a dedicated, line-by-line lesson — including the two that a prior draft of this course left completely uncovered: document_service.py (chunking, embeddings, ChromaDB writes) and rag_graph_service.py (the one real LangGraph StateGraph in this codebase).

What You'll Learn#

• Every function in openai_service.py: chat completion, email analysis, RAG answers, retrieval evaluation
• The token/cost estimation math behind every ToolCall row
• The three-agent email pipeline: analysis → reply → review, and how it differs from a LangGraph workflow
• The character-based chunking algorithm, embedding calls, and ChromaDB writes behind document ingestion
• The Corrective RAG StateGraph: four nodes, one conditional edge, and why the branch exists at all

Prerequisites#

• Module 2: Data Layer — services consume the models and schemas from that module directly

Lessons in This Module#

Learning Outcomes#

By the end of this module you can:

• List all six system prompts in this codebase and which ones expect JSON back
• Explain the character-based chunking algorithm's overlap math by hand
• Draw the Corrective RAG graph from memory: nodes, edges, and the one conditional branch
• State precisely why the email workflow is not LangGraph, and what the actual difference is in code

Start Module → Back to Module 2