SkyMate

Home
Chat
Form
Form

Inspiration

Travel today offers endless possibilities, but that abundance creates uncertainty. We were inspired by the idea that users don’t need more options, they need better decisions. Skyscanner’s challenge pushed us to rethink travel planning as an intent-driven experience, not just a search problem.

What it does

SkyMate is an AI-powered travel assistant that understands user intent and helps them move from uncertainty to confident travel decisions.

It allows users to:

Find trips based on real constraints (dates, budget, people)
Get personalized recommendations tailored to their preferences
Discover destinations through an inspiration flow

Instead of just responding to prompts, SkyMate interacts with users through dynamic forms and conversations, gathering context to deliver smarter, more relevant results powered by Skyscanner data.

How we built it

We built SkyMate as a full-stack web application:

Frontend: Next.js + Tailwind CSS for a responsive, chat-based UI
Backend: FastAPI (Python) to orchestrate logic and AI interaction
AI Layer: Local LLM (Gemma via Ollama) integrated with LangChain agents
Agentic AI: The model can decide when to use tools (e.g. Skyscanner API functions)
RAG-style context: We enrich prompts with structured data from user inputs (forms + chat)
Tooling: Custom functions for flight search, airport resolution, and user context

The system combines conversational AI, structured inputs, and real-time data into a unified decision engine.

Challenges we ran into

Designing a system that understands intent, not just prompts
Integrating tool-calling AI reliably with real APIs
Structuring user inputs (forms + chat) into meaningful context for the model
Managing conversation state across frontend and backend
Running and optimizing a local LLM within time and resource constraints

Accomplishments that we're proud of

Building a fully functional agentic AI system in a hackathon setting
Successfully integrating real Skyscanner data into AI responses
Creating an experience that combines forms + chat, not just a chatbot
Designing a product that focuses on decision-making, not just information retrieval
Delivering a clean, intuitive UI aligned with real product standards

What we learned

We learned deeply about AI systems end-to-end, from basic LLM usage to building a full Agentic AI architecture
How to design and implement agentic workflows, where the model decides when to use tools and external data
How to combine real-time API data (Skyscanner) with LLM reasoning to generate grounded, useful responses
How to run and integrate local AI models (Gemma via Ollama) inside a web application
The importance of structuring context (forms + chat) to significantly improve response quality
That building AI products is as much about product design and user experience as it is about the model itself

What's next for SkyMate

Improve how we collect and structure user intent to generate even more accurate responses
Add more interactive and intuitive UI elements to make the experience smoother and more engaging
Expand the system to gather richer context and continuously improve response quality
Integrate additional travel features (hotels, experiences, dynamic recommendations)
Enhance the agent with better reasoning and tool usage capabilities