Loading
Ace Intelligence Systems
Preparing a calmer, clearer view of your automation workspace.
Aligning workflows
Loading
Preparing a calmer, clearer view of your automation workspace.
Ace Intelligence
An intelligent chatbot system built using n8n workflow automation that handles multi-modal inputs from Telegram — text, audio, images, and documents.
This project leverages Retrieval-Augmented Generation (RAG) techniques to enhance AI responses with relevant information retrieved from a knowledge base. It processes multi-modal inputs by converting them into embeddings, storing them in a vector database (Milvus), and providing context-aware responses using advanced language models. The system maintains conversation memory for coherent interactions and supports various document formats for knowledge ingestion.
Watch the full demo on YouTube to see the Multi-Modal RAG Agent in action — processing text, images, audio, and documents via Telegram.
Multi-Modal Input Handling: Supports text, audio, images, and documents from Telegram messages. Vector Database Integration: Uses Milvus for efficient similarity search and retrieval. Advanced Embeddings: Employs Cohere's multilingual embeddings for accurate semantic understanding. Conversational Memory: Maintains context across interactions for natural conversations. Document Processing: Automatically extracts and chunks content from PDFs and other files. Real-time Responses: Provides instant replies via Telegram bot interface. Scalable Architecture: Built on n8n's workflow engine for easy customization. Webhook Support: Integrated with ngrok for external API access.
Telegram Integration: Receives messages and media from users via Telegram Bot API. Data Processing Pipeline: Extracts text from various formats (PDF, audio transcription, image OCR). Embedding Generation: Converts processed content into vector embeddings using Cohere. Vector Storage: Stores embeddings in Milvus vector database for fast retrieval. Retrieval System: Performs similarity search to find relevant context for user queries. Language Model: Uses GPT-4o-mini to generate responses based on retrieved information. Response Delivery: Sends formatted replies back through Telegram. The workflow is orchestrated through n8n, providing a visual interface for monitoring and modifying the agent's behavior.
Customer Support: Provide instant, knowledgeable responses based on company documentation. Educational Assistant: Answer questions using uploaded textbooks, research papers, or course materials. Research Helper: Retrieve and summarize information from scientific documents. Personal Knowledge Base: Build a searchable database of personal notes, articles, and media. Content Creation: Generate responses informed by reference materials and style guides. Multilingual Support: Handle queries in multiple languages with multilingual embeddings.
n8n for workflow orchestration. Milvus for vector database. Cohere for multilingual embeddings. GPT-4o-mini for language model. Telegram Bot API for messaging. Docker for deployment. ngrok for webhook tunneling.
n8n | Milvus | Cohere | GPT-4o-mini | Telegram | Docker | ngrokExplore the full source code, docker-compose setup, and n8n workflow JSON on GitHub.
https://github.com/OMCHOKSI108/AI-AUTOMATION-WORKFLOWS/tree/main/MULTI_MODEL_RAG_AGENT