Travliaq turns one sentence ("Barcelona, 4 days, €700, as a couple") into a complete, personalised trip plan. Instead of one giant prompt, the work is split across a crew of specialised AI agents, each with its own role, tools and output format.
My role
Product, architecture and end-to-end build, from the agent system to the web app.
What I built
8 CrewAI agents covering orchestration, traveller persona analysis, destination suggestions, flights, accommodation, activities, itinerary validation and formatting
A persona inference engine that only recommends above 70% confidence, so suggestions stay relevant
Real travel data through the Amadeus API and dedicated data collectors exposed to the agents through an MCP server
A FastAPI backend with structured logging and error handling, and a React and TypeScript front end
Engineering choices
Agents and tasks declared in YAML, so behaviour can be tuned without touching code
A validation agent checks budget and date consistency before anything reaches the user
Fast inference on Groq (Llama 3.3 70B), with Azure OpenAI as a second provider
Stack
Python, CrewAI, FastAPI, MCP, Groq, Azure OpenAI, Amadeus API, Supabase, React, TypeScript
0 likes · 0 views · Nov 10, 2026
Conversational SaaS where 8 specialised AI agents turn a single request into a personalised, ready-to-book trip plan.
