sidequest

sidequest interface/dashboard

Inspiration

Planning travel is broken. As college students, we’ve experienced how fragmented and time-consuming it is to turn inspiration into an actual booked trip. While 84% of Gen Z use social media for travel inspiration, travelers still spend 8.5+ hours consuming content before they ever book. Platforms like TikTok and Instagram fuel inspiration, but converting that inspiration into flights, hotels, and a coherent itinerary requires juggling dozens of tabs, apps, and price comparisons.

The problem isn’t a lack of information, it’s decision fatigue. We wanted to build something that doesn’t just suggest ideas, but actually executes on them.

What it does

sidequest is an agentic AI travel assistant that plans, refines, and executes trips end-to-end.

Users describe their trip in natural language, or paste in social media travel videos, and sidequest:

Generates a structured, day-by-day itinerary
Estimates total trip cost in real time
Allows users to customize activities and hotel/flight options
Initiates booking through links for flights, hotels, and activities

Instead of searching and comparing manually, users delegate intent to an AI agent that decides on their behalf.

How we built it

sidequest is built as a web-based agentic AI system optimized for fast iteration and real-world execution.

Frontend:

Next.js (App Router)
Tailwind CSS + shadcn/ui
Responsive layout with a chat interface + live itinerary sidebar

Backend:

Next.js API routes
Zod schemas to enforce structured AI outputs
Stateful itinerary management for iterative planning

AI & Agent Logic:

OpenAI models used as a goal-driven agent, not a chatbot
The agent follows a ReAct-style loop:
- Reason about user intent
- Call tools (planning, pricing, discovery)
- Update itinerary state
- Re-plan based on user feedback

Execution Layer:

Google Places for activity discovery
Deep-linked booking flows (flights, hotels, activities)
Real-time cost estimation to support decision-making

Overshoot:

We used Overshoot to extract locations and geographic signals from TikTok and Instagram travel videos
Detected places are aggregated across videos and passed into our AI agent as hard planning constraints
This lets sidequest turn social media inspiration directly into structured, bookable itineraries, closing the loop from discovery → planning → execution

Challenges we ran into

Alternating our API keys when testing without maxing out our usage.

Enforcing structured AI output: LLMs naturally produce prose, so we had to strictly constrain responses to valid JSON and implement retries and validation.

Balancing realism with hackathon constraints: True in-app bookings require enterprise-level partnerships, so we designed flows that are demo-realistic while still honest about scope.

Maintaining state across revisions: Allowing users to iteratively modify plans required careful state management so the agent didn’t “start over” each time.

Time pressure: Building an end-to-end product in under 24 hours required prioritization.