graph TD %%{init: {'theme': 'mc','layout': 'elk'}}%% ChatInput-larrt[<div><img alt="logo" src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Input] style ChatInput-larrt stroke:#a170ff GoogleGenerativeAIModel-hrflu[Gemini] style GoogleGenerativeAIModel-hrflu stroke:#a170ff GoogleGenerativeAIEmbeddings-fnjt0[Google Generative AI Embeddings] style GoogleGenerativeAIEmbeddings-fnjt0 stroke:#a170ff APIRequest-78xr1[API Request] style APIRequest-78xr1 stroke:#a170ff TelegramMessage-zvrp9[Telegram Message] style TelegramMessage-zvrp9 stroke:#a170ff ChatOutput-16z2g[<div><img alt="logo" src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Output] style ChatOutput-16z2g stroke:#a170ff ChatOutput-0vfib[<div><img alt="logo" src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Output2] style ChatOutput-0vfib stroke:#a170ff ChatOutput-7h910[<div><img alt="logo" src="/_astro/messages-square.BaSDmT6g.svg" style="height: 20px !important;width: 20px !important"/></div>Chat Output3] style ChatOutput-7h910 stroke:#a170ff GoogleGenerativeAIEmbeddings-mn2ew[Google Generative AI Embeddings2] style GoogleGenerativeAIEmbeddings-mn2ew stroke:#a170ff SplitText-xm92m[<div><img alt="logo" src="/_astro/scissors-line-dashed.CajsPhTx.svg" style="height: 20px !important;width: 20px !important"/></div>Split Text] style SplitText-xm92m stroke:#a170ff GDriveFilesComponent-q2n9y[<div><img alt="logo" src="/_astro/google_drive.wKmDsV2c.svg" style="height: 20px !important;width: 20px !important"/></div>Google Drive File Manager] style GDriveFilesComponent-q2n9y stroke:#a170ff ParseData-dpd2k[<div><img alt="logo" src="/_astro/braces.Djq0PW4_.svg" style="height: 20px !important;width: 20px !important"/></div>Parse Data] style ParseData-dpd2k stroke:#a170ff AdvancedAgent-zbpzj[Agent4] style AdvancedAgent-zbpzj stroke:#a170ff AdvancedAgent-aazz1[Agent] style AdvancedAgent-aazz1 stroke:#a170ff AdvancedAgent-hs2r4[Agent5] style AdvancedAgent-hs2r4 stroke:#a170ff ParseData-2g4fn[<div><img alt="logo" src="/_astro/braces.Djq0PW4_.svg" style="height: 20px !important;width: 20px !important"/></div>Parse Data2] style ParseData-2g4fn stroke:#a170ff ParseData-mrvuk[<div><img alt="logo" src="/_astro/braces.Djq0PW4_.svg" style="height: 20px !important;width: 20px !important"/></div>Parse Data3] style ParseData-mrvuk stroke:#a170ff ParseData-z8755[<div><img alt="logo" src="/_astro/braces.Djq0PW4_.svg" style="height: 20px !important;width: 20px !important"/></div>Parse Data4] style ParseData-z8755 stroke:#a170ff DeepseekModel-7c7w4[Deepseek2] style DeepseekModel-7c7w4 stroke:#a170ff DeepseekModel-juyng[Deepseek3] style DeepseekModel-juyng stroke:#a170ff DeepseekModel-ova3h[Deepseek4] style DeepseekModel-ova3h stroke:#a170ff SmartRouter-kldld[Agent Router] style SmartRouter-kldld stroke:#a170ff RetrieverTool-65vrx[<div><img alt="logo" src="/_astro/langchain-icon.BXtvU_nA.svg" style="height: 20px !important;width: 20px !important"/></div>RetrieverTool] style RetrieverTool-65vrx stroke:#a170ff AstraDB-bdkzn[Astra DB] style AstraDB-bdkzn stroke:#a170ff AstraDB-xlv27[Astra DB2] style AstraDB-xlv27 stroke:#a170ff ParseData-dpd2k -.- SplitText-xm92m linkStyle 0 stroke:#a170ff GDriveFilesComponent-q2n9y -.- ParseData-dpd2k linkStyle 1 stroke:#a170ff AdvancedAgent-zbpzj -.- ChatOutput-16z2g linkStyle 2 stroke:#a170ff AdvancedAgent-aazz1 -.- ChatOutput-7h910 linkStyle 3 stroke:#a170ff APIRequest-78xr1 -.- AdvancedAgent-aazz1 linkStyle 4 stroke:#a170ff AdvancedAgent-hs2r4 -.- ChatOutput-0vfib linkStyle 5 stroke:#a170ff TelegramMessage-zvrp9 -.- AdvancedAgent-hs2r4 linkStyle 6 stroke:#a170ff ParseData-2g4fn -.- AdvancedAgent-zbpzj linkStyle 7 stroke:#a170ff ParseData-mrvuk -.- AdvancedAgent-aazz1 linkStyle 8 stroke:#a170ff ParseData-z8755 -.- AdvancedAgent-hs2r4 linkStyle 9 stroke:#a170ff DeepseekModel-7c7w4 -.- AdvancedAgent-zbpzj linkStyle 10 stroke:#a170ff DeepseekModel-juyng -.- AdvancedAgent-aazz1 linkStyle 11 stroke:#a170ff DeepseekModel-ova3h -.- AdvancedAgent-hs2r4 linkStyle 12 stroke:#a170ff ChatInput-larrt -.- SmartRouter-kldld linkStyle 13 stroke:#a170ff GoogleGenerativeAIModel-hrflu -.- SmartRouter-kldld linkStyle 14 stroke:#a170ff SmartRouter-kldld -.- ParseData-2g4fn linkStyle 15 stroke:#a170ff SmartRouter-kldld -.- ParseData-mrvuk linkStyle 16 stroke:#a170ff SmartRouter-kldld -.- ParseData-z8755 linkStyle 17 stroke:#a170ff RetrieverTool-65vrx -.- AdvancedAgent-zbpzj linkStyle 18 stroke:#a170ff SplitText-xm92m -.- AstraDB-bdkzn linkStyle 19 stroke:#a170ff GoogleGenerativeAIEmbeddings-mn2ew -.- AstraDB-bdkzn linkStyle 20 stroke:#a170ff AstraDB-xlv27 -.- RetrieverTool-65vrx linkStyle 21 stroke:#a170ff GoogleGenerativeAIEmbeddings-fnjt0 -.- AstraDB-xlv27 linkStyle 22 stroke:#a170ff

📄 Ecommerce Customer Service – Automated Query Handling

🧩 Overview

The workflow automates the entire customer‑service experience for an e‑commerce platform.
When a user submits a question in the chat, the system classifies the request into one of three intent levels:

Level 1 – Information‑only queries (e.g., company background).
Level 2 – Operational queries (e.g., order status, inventory).
Level 3 – Escalation queries (e.g., returns, refund requests).

For each intent the system selects the appropriate agent, which either retrieves knowledge from a vector store, calls an external API, or forwards the request to a human employee via Telegram. The final response is displayed back to the user in the chat interface.

⚙️ Main Features

Intent routing – AI‑powered classifier directs user messages to the correct agent.
Vector‑based knowledge base – documents are chunked, embedded, and indexed in Astra DB for semantic search.
LLM integration – Deepseek or Gemini models generate natural‑language answers.
Dynamic tool usage – Agents call external APIs (order status) or send notifications (Telegram) as needed.
Multi‑step agent logic – Each agent can perform several actions, including reasoning, retrieval, and tool execution, before delivering a final reply.
Real‑time chat output – Responses are streamed to the user as they are produced.

🔄 Workflow Steps

#	Component Name	Role in the Workflow	Key Inputs	Key Outputs
1	Google Drive File Manager	Pulls raw documents from Google Drive	File selection / ID	Binary file content
2	Parse Data (File)	Converts binary file to plain text	File content	Text string
3	Split Text	Breaks large text into 1 000‑char chunks with 200‑char overlap	Full text	Chunked text array
4	Google Generative AI Embeddings	Generates high‑dimensional vectors for each chunk	Text chunks	Embedding vectors
5	Astra DB – Add Documents	Stores embeddings and metadata in a vector store	Embeddings, metadata	Success flag
6	Astra DB – Base Retriever	Builds a retriever for semantic search	Vector store	Retriever instance
7	Retrieval Tool	Exposes the retriever as a callable tool	Retriever	Tool object
8	Deepseek Model (Level 1)	Provides the language model for the Level 1 agent	Model parameters	LLM instance
9	Deepseek Model (Level 2)	Provides the language model for the Level 2 agent	Model parameters	LLM instance
10	Deepseek Model (Level 3)	Provides the language model for the Level 3 agent	Model parameters	LLM instance
11	Chat Input	Receives user messages	Message text	Message object
12	Smart Router	Classifies message into `nivel1`, `nivel2`, or `nivel3`	User message, LLM	Intent output data
13	Parse Data (Level 1)	Formats intent data for the Level 1 agent	Intent data	Text string
14	Advanced Agent – Level 1	Answers using the Retrieval Tool	Prompt, Retrieval Tool, LLM	Text response
15	Chat Output – Level 1	Displays Level 1 answer	Response text	Rendered chat message
16	Parse Data (Level 2)	Formats intent data for the Level 2 agent	Intent data	Text string
17	API Request	Calls external e‑commerce API (e.g., order status)	URL, headers, body	API response
18	Advanced Agent – Level 2	Uses API result to compose answer	Prompt, API response, LLM	Text response
19	Chat Output – Level 2	Displays Level 2 answer	Response text	Rendered chat message
20	Parse Data (Level 3)	Formats intent data for the Level 3 agent	Intent data	Text string
21	Telegram Message	Sends a notification to a human agent	Message text, chat ID	Telegram API response
22	Advanced Agent – Level 3	Calls Telegram tool and confirms escalation	Prompt, Telegram tool, LLM	Confirmation text
23	Chat Output – Level 3	Displays Level 3 confirmation	Response text	Rendered chat message

🧠 Notes

File loading is performed once at startup; subsequent queries use the pre‑built vector store.
Retrieval Tool is only required for Level 1 agents; Level 2 agents rely on the external API.
Deepseek Models are instantiated per intent level; this allows independent tuning and cost control.
API Request must include a valid authentication token; otherwise the Level 2 agent will return an error message.
Telegram Message requires the bot token and chat ID to be configured; missing credentials will prevent escalation notifications.
Intent thresholds are set at 0.7 confidence; lower thresholds can increase routing errors, higher thresholds may miss ambiguous queries.
Memory handling is disabled in the agents to keep responses stateless; enable if conversational context is needed.
Streaming is optional; enabling it provides real‑time feedback to the user but increases resource usage.

This documentation summarizes the logical flow, component responsibilities, and key data exchanges within the e‑commerce customer‑service automation pipeline.