SkillUp Operations Support Bot¶

An AI-powered support assistant that helps SkillUp Bootcamp team members (Program Officers, Trainers, and Site Assistants) navigate daily operations, safeguarding protocols, and training delivery through WhatsApp/SMS and Telegram messaging interfaces. The bot uses RAG (Retrieval-Augmented Generation) to provide accurate, context-aware guidance based on uploaded operational handbooks and documentation.

Purpose¶

No business context provided yet — add a context.md to enrich this documentation.

Based on the workflow configuration, this bot appears to serve the SkillUp Bootcamp program, which supports Out-of-School Youth (OoSY) in Kenya seeking entrepreneurial skills. The bot helps field staff with:

Daily bootcamp operations and procedures
Child protection and safeguarding protocols
Retention and activation strategies
Vendor and materials management
Team dynamics and professional conduct
Youth engagement and support

The bot embodies Educate!'s cultural values and maintains strict safeguarding standards while providing practical, actionable guidance to ensure program quality and participant safety.

How It Works¶

Document Upload: Staff upload operational handbooks, procedures, and training materials via a web form (PDF/CSV files)
Knowledge Processing: Documents are processed, split into chunks, and converted to embeddings using OpenAI
Vector Storage: Document embeddings are stored in an in-memory vector database for quick retrieval
Message Reception: The bot receives questions via WhatsApp/SMS (Twilio) or Telegram
Context Retrieval: When a question arrives, the system searches the knowledge base for relevant information
AI Response: An OpenAI GPT-4o-mini model generates contextual responses using the retrieved knowledge and specialized prompts
Memory Management: Conversation history is maintained per user session (20-message window)
Response Delivery: Answers are sent back through the same messaging channel

Workflow Diagram¶

graph TD
    A[Upload your file here] --> B[Insert Data to Store]
    B --> C[Default Data Loader]
    C --> D[Embeddings OpenAI]
    D --> E[Query Data Tool]

    F[Telegram Trigger] --> G[AI Agent]
    H[Twilio Trigger] --> G

    G --> I[Send a text message]
    G --> J[Send an SMS/MMS/WhatsApp message]

    K[OpenAI Chat Model] --> G
    L[Simple Memory] --> G
    E --> G

    D --> B
    D --> E

Trigger¶

This workflow has three triggers:

Form Trigger: Web form for uploading documents (inactive/archived)
Telegram Trigger: Receives messages from Telegram bot
Twilio Trigger: Receives WhatsApp/SMS messages via Twilio webhook

Nodes Used¶

Node Type	Purpose
Form Trigger	Provides web interface for document uploads
Default Data Loader	Processes uploaded binary files (PDF, CSV)
Embeddings OpenAI	Converts documents to vector embeddings
Vector Store In Memory	Stores and retrieves document embeddings
AI Agent	Orchestrates the conversation flow and response generation
OpenAI Chat Model	Generates responses using GPT-4o-mini
Simple Memory	Maintains conversation context per user
Telegram Trigger	Receives Telegram messages
Twilio Trigger	Receives WhatsApp/SMS messages
Telegram	Sends responses via Telegram
Twilio	Sends responses via WhatsApp/SMS

External Services & Credentials Required¶

OpenAI¶

Purpose: Document embeddings and chat completions
Credential: OpenAI API key
Models Used: text-embedding-ada-002, gpt-4o-mini

Telegram¶

Purpose: Bot messaging interface
Credential: Telegram Bot Token
Setup: Create bot via @BotFather

Twilio¶

Purpose: WhatsApp/SMS messaging
Credential: Twilio Account SID and Auth Token
Setup: Configure webhook URL for inbound messages

Environment Variables¶

No specific environment variables are defined in this workflow. All configuration is handled through n8n credentials and node parameters.

Data Flow¶

Input¶

Documents: PDF and CSV files containing operational procedures, handbooks, and training materials
Messages: Text messages from staff via Telegram or WhatsApp/SMS asking operational questions

Processing¶

Documents are chunked, embedded, and stored in vector database
Incoming messages are matched against stored knowledge
AI agent generates responses using retrieved context and specialized prompts

Output¶

Responses: Contextual guidance and answers sent back through the same messaging channel
Format: Plain text messages with actionable advice, procedures, and escalation guidance

Error Handling¶

The workflow does not implement explicit error handling nodes. Error management relies on n8n's default behavior and individual node error handling capabilities.

Known Limitations¶

Memory Storage: Uses in-memory vector storage, which doesn't persist between workflow restarts
File Types: Limited to PDF and CSV uploads only
Archived Status: The workflow is currently archived and inactive
No Persistence: Document knowledge base needs to be rebuilt after system restarts
Single Session: Memory is maintained per chat ID but limited to 20 messages

No related workflows are specified in the current context.

Setup Instructions¶

Import Workflow: Import the JSON into your n8n instance
Configure Credentials:
- Set up OpenAI API credentials with access to embeddings and chat models
- Create Telegram bot and configure bot token
- Set up Twilio account with WhatsApp/SMS capabilities
Upload Knowledge Base:
- Activate the workflow
- Use the form trigger to upload operational handbooks and procedures
- Ensure documents are in PDF or CSV format
Configure Messaging:
- Set up Telegram webhook URL in bot settings
- Configure Twilio webhook to point to the Twilio Trigger endpoint
- Test message flow with sample questions
Customize AI Prompts:
- Review and adjust the system prompt in the AI Agent node
- Modify response style and escalation procedures as needed
- Update knowledge base tool description if required
Activate Workflow: Enable the workflow to start receiving and responding to messages

Note: The workflow is currently archived. You'll need to unarchive it before activation and testing.