n8n

This project is an AI-powered ticket triage system that automatically analyses incoming Autotask tickets, classifies them, and routes them to the correct queue, support agent, service contract, issue type, and sub-issue type. The goal is to reduce manual ticket handling, improve response time, and make ticket assignment more consistent across support teams.

This project is an AI-powered trading analysis platform that analyses the market and generates trading signals such as LONG, SHORT, or STANDBY based on market conditions. The system connects to user accounts through the MEXC API, allowing users to link their exchange account and use the generated signals within their own trading setup. A PWA version is also configured for users who need a mobile app-like experience without installing a native application. The platform includes simulation and paper trading features, allowing users to test signals safely, compare different strategies, and identify which setup best suits their needs before using real funds. Each signal also includes AI reasoning, giving users a clear explanation and justification behind why the agent produced that trading decision.

This project is an AI-powered social media automation agent that connects to a user’s LinkedIn and X accounts to help create posts, generate AI images, comment, repost, and react to content based on the user’s settings. Users can define impersonation tones so the agent writes in a style that sounds like them, making posts and interactions feel more personal and natural. The system also supports one-click re-authentication when account access tokens expire, making it easy for users to reconnect their accounts. Users can configure the agent’s active time and choose which actions it is allowed to perform, giving them control over when and how the agent engages online.

Role: AI Developer - retrieval pipeline, agent orchestration, and injection defence. Front-end built by other team members. The system. Customer support questions are answered from a Pinecone vector database holding the company's domain knowledge. The orchestration runs entirely in n8n, with four GPT-5.6 Luna agents dividing the work. Luna was chosen for its speed and cost profile, a multi-agent pipeline means several model calls per user message, so a fast, inexpensive model keeps both latency and per-conversation cost viable at support volume.