Jento - AI Trip Planner

Inspiration

Jento started from a problem we experienced ourselves — planning a trip can sometimes feel like a project on its own. We found ourselves jumping between Instagram reels, YouTube videos, Google Maps, booking websites, screenshots, and notes just to plan a few days away.

We wanted to build something that could take all those scattered ideas and turn them into a trip that actually feels personal. That’s how Jento came about: an AI travel companion that helps you go from “Where should we go?” to “Okay, our trip is planned!” without all the hassle.

What it does

Jento is an AI-powered travel companion that helps you plan and manage trips in one place.

Enter your destination, dates, budget, interests, and preferences, and Jento creates a personalised itinerary you can easily customise. You can also plan quick weekend getaways, discover new places, or find places based on your interests.

Jento also helps you:

Estimate trip expenses
Generate activity-based packing lists
Store bookings and travel documents
Share and edit trips with your companions
Split your expenses.

How we built it

We started by breaking down the travel-planning journey and identifying where people spend the most time and effort. From there, we designed and tested the experience, then built the product around AI-generated itineraries.

Early on we learned something important: pure LLM itineraries aren’t enough. A model is great at interpreting “romantic weekend under ₹85,000” and suggesting places that feel right — but it struggles with opening hours, travel times, meal windows, and a hard budget. So we built Jento around a hybrid planning engine: the LLM proposes, a deterministic solver enforces the constraints.

That pipeline works in stages:

Propose — Gemini turns the user’s trip into prioritised, grounded stop candidates (with place IDs from Google Maps).
Ground — We resolve each stop against live Places data: opening hours, open/closed status, price signals, and attributes like vegetarian-friendly or pet-friendly.
Substitute — Closed, off-diet, or otherwise invalid venues are swapped for real nearby alternatives.
Optimise (two stages) — Per day, an exact dynamic program schedules stops under opening hours, travel time, meal windows, and pace. Across the trip, a local search moves stops between days (e.g. a museum closed on Mondays) while keeping multi-city routes coherent.
Budget repair — If the plan exceeds the user’s budget, we greedily swap hotels and restaurants, then drop low-value paid activities until it fits — or explain clearly why it can’t.
Verify — An independent checker re-validates the final plan so we don’t trust the solver blindly.

We gradually added features that solve the smaller problems around a trip too — estimating expenses, packing lists from the itinerary, travel documents, splitting expenses, and weekend plans. These came from thinking about what happens before, during, and around the trip, not just the itinerary itself.

Our goal throughout was to keep the experience simple:

Tell Jento what you want, get a plan, and make it yours.

Challenges we ran into

The single hardest problem wasn’t the AI itself, but making its output actually work in the real world.

Gemini would sometimes suggest a museum on Monday when it’s closed, a “short walk” between stops 8 km apart, or a restaurant that closed two years ago — and users would find out on day one, standing in front of a locked door in a foreign city. Budgets were another trap: the model might happily recommend 5-star stays that quietly double the target.

That gap between “sounds plausible” and “logistically correct” is where most trip planners quietly die.

We tackled it with a hybrid, server-side trust layer:

Strict schemas and prompts (Zod + Gemini) — every itinerary is structured before it reaches the database.
Google Maps grounding — every place is resolved against real Place IDs.
Constraint-aware planner — accounts for opening hours, travel times, meal windows, pace, dietary/infant/pet needs, and budget.
Two-stage scheduling — exact per-day dynamic programming (DP) + cross-day local search, instead of relying on the LLM to “figure out” logistics.
Budget allocation and repair — grounded in real place pricing and travel-cost floors.
Schedule recomputation — plans are recalculated when users edit, reorder, or add ideas.
Disconnect-safe generation — plans survive mobile network drop-offs and interrupted generation.

This ensures Jento doesn’t just chat about trips — it produces itineraries you can actually follow.

Accomplishments that we're proud of

We're especially proud of creating an experience where AI isn't just a chatbot — it actually helps users create, customise, and manage an entire trip in one place.

We've built more than just an AI itinerary generator by focusing on a complete travel planning system where features work together. One of our biggest accomplishments is the hybrid engine itself: Jento can take destination, dates, budget, interests, and preferences and produce a feasible day-by-day plan — not a generic list of recommendations that ignores time, distance, or money.

This gives Jento an edge because we're not treating AI as a standalone chatbot; we're using it as the intelligence layer across the entire travel-planning experience, with formal verification for the parts where language models fail.

What we learned

Building Jento taught us that a good AI product needs more than a good prompt. We had to constantly move between designing, building, testing, and refining the experience.

We learned two key technical lessons:

1. Grounding beats hallucination

An AI inventing a restaurant or hotel can quickly break user trust.

We use Gemini’s Google Maps grounding to verify recommendations and attach real Google Place IDs, while using trip coordinates to keep recommendations geographically relevant.

2. LLMs need structure, not just flexibility

Free-form chat works well for discovery, but itineraries need consistency.

We used Zod and Gemini to ensure every itinerary element follows a predictable structure before reaching our database.

3. Linguistic flair needs mathematical verification

Research and our own testing showed the same pattern: standalone models invent beautiful trips that break under real constraints. The breakthrough was treating itinerary generation as a combinatorial optimisation problem — LLM as translator and proposer, solvers for schedules and budget, verifier for hard guarantees.

We also learned that AI doesn't have to do everything for the user. For Jento, people want a starting point, not a fixed plan. AI creates the day-by-day itinerary; the user can still swap a place, change the pace, or add something they discovered themselves — and the planner re-checks feasibility when they do.

What's next for Jento — AI Trip Planner

We're just getting started.

Our next focus is to make Jento smarter, more personalised, and more useful throughout the entire travel journey.

We want Jento to:

Better understand individual travel styles and preferences
Improve recommendations based on real user behaviour and feedback
Become more personalised over time
Eventually become something you can rely on from the moment you start thinking about a trip until you come back home

Jento isn't just about planning your next trip. We're building an AI travel companion for the entire journey.