Skip to content

Latest commit

 

History

2 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 

Repository files navigation

✈️ TripPilot — Multi-Agent AI Travel Planning Product

From Messy Multi-Tab Planning to One Guided AI Flow | 6 Specialist Agents | Human-Controlled Bookings

Status Agents Framework Research


🧭 The Story — Why TripPilot Exists

In March 2025, Skift reported that Google had indefinitely shelved its AI trip planner.

The biggest technology company in the world looked at the problem of AI-powered travel planning and said: "We can't solve this."

That's not a dead end. That's a product opportunity.

Here's what the current travel planning experience looks like for most people:

Tab 1: Google Flights (price check)
Tab 2: MakeMyTrip (booking)
Tab 3: TripAdvisor (reviews)
Tab 4: Google Maps (distances)
Tab 5: Currency converter
Tab 6: Weather forecast
Tab 7: Visa requirements
Tab 8: Hotel comparison
...12 tabs later, decision fatigue sets in

Real users. Real frustration. Real reviews:

"The itinerary designed by MMT was extremely poorly planned. Hotel checkout was at 11 AM but our return flight was at 11:30 PM — we were forced to sit at the airport for the entire day." — ConsumerAffairs, Feb 2026

"They have some trick coding. Whenever you click back to adjust something, the original discount is gone." — ConsumerAffairs, Nov 2024

"Many smaller or independent hotels I see on Google end up not being connected to the pricing databases." — Rick Steves Travel Forum, 2024

TripPilot's thesis: The problem isn't information. The problem is that no platform connects research, budget, booking, and context into a single intelligent flow — with AI that actually understands the relationships between all the moving parts.


🎯 The Job To Be Done

"Help me confidently plan and book a trip that fits my time, budget and preferences — without stress."

That's it. Everything in TripPilot's design flows from this one sentence.


🤖 The 6-Agent Architecture

USER INPUTS
Dates · Budget · Preferences · Constraints · Companions
                    ↓
            ORCHESTRATOR
     Controls which agent runs, in what order
     Manages guardrails and human control gates
                    ↓
    ┌───────────────────────────────────────┐
    │                                       │
    │  🔍 RESEARCH AGENT                    │
    │  Fetches flights, hotels, activities  │
    │  Clean untracked sessions → real      │
    │  prices (no dynamic pricing bias)     │
    │  Aggregates OTAs + direct APIs        │
    │                                       │
    │  💰 BUDGET AGENT                      │
    │  Locks TRUE total cost upfront        │
    │  Fees + taxes + transfers + baggage   │
    │  No hidden charges at checkout        │
    │                                       │
    │  📅 BOOKING AGENT ← HUMAN GATE 🔴    │
    │  User reviews & approves before       │
    │  any booking is confirmed             │
    │  Live confirmation verification       │
    │                                       │
    │  🗺️ ITINERARY AGENT                  │
    │  Cross-checks checkout vs flight      │
    │  times, transfer durations, crowds    │
    │  Builds conflict-free schedule        │
    │                                       │
    │  📍 CONTEXT AGENT                     │
    │  Hyperlocal recommendations           │
    │  Real addresses, not vague areas      │
    │  Weather · Safety · Local events      │
    │                                       │
    │  🔔 MONITOR AGENT                     │
    │  Pre-trip alerts (24hr before)        │
    │  Price drops · Cancellation risks     │
    │  Rebooking suggestions                │
    │                                       │
    └───────────────────────────────────────┘
                    ↓
            OUTPUTS TO USER
    Itinerary · Booking links · Budget dashboard · Alerts

🎨 Design Principles

1. Human Control for Risky Steps

The Booking Agent never books without explicit user approval. This is hardcoded — not configurable. One accidental booking on a non-refundable hotel is a trust-destroying experience. The AI recommends. The human confirms.

2. Automation for Repetitive Steps

Searching 20 flight combinations, comparing 15 hotels, calculating true total costs including transfer fees — all of this happens automatically. This is where AI earns its place.

3. Always Explainable

Every recommendation comes with a "Why this?" rationale. "I chose this hotel because it's 8 minutes from your first activity, within your budget, and has free cancellation until 48 hours before arrival." The AI shows its work.

4. True Total Cost — Always

What other platforms show:    ₹40
What TripPilot shows:         ₹40 base
                              + ₹180 convenience fee
                              + ₹240 GST
                              + ₹140 airport transfer
                              = ₹600 TRUE TOTAL

🔬 The Research Behind It — 10 Validated Pain Points

This product was grounded in real user pain, not assumptions. Every pain point sourced from real consumer reviews.

# App Pain Point TripPilot Solution
M1 MakeMyTrip Hotel checkout 11am, flight 11:30pm — 12 hours stranded Itinerary Agent cross-checks all time conflicts
M2 MakeMyTrip Discounts disappear when you click back Budget Agent locks price at point of search
M3 MakeMyTrip Hotel had no record of confirmed booking Booking Agent verifies directly with hotel API
M4 MakeMyTrip ₹40 shown, ₹600 charged at checkout True total cost shown upfront — always
T1 TripAdvisor Last-minute hotel cancellation, no resolution Monitor Agent tracks booking status pre-trip
T2 TripAdvisor Restaurant recommendation had wrong address Context Agent delivers precise hyperlocal data
T3 TripAdvisor Wrong tour booked — non-refundable, no help Human control gate before every booking
G1 Google Flights Prices inflate based on browsing history Research Agent fetches in clean untracked sessions
G2 Google Flights Independent hotels not in database Research Agent aggregates across OTAs + direct APIs
G3 Google AI trip planner shelved indefinitely TripPilot fills the market gap Google left open

🔄 ERRC Framework — How TripPilot Changes the Game

Action What We Do
Eliminate Tab switching · Repeated data entry · Hidden fee surprises · Booking confirmation anxiety
Reduce Time-to-plan · Decision fatigue · Price comparison effort · Post-booking uncertainty
Raise Price transparency · Booking confidence · Itinerary coherence · "Why this?" explainability
Create Single guided flow: plan → optimize → book · True total cost · Pre-trip monitoring · Budget reverse-search

💡 SCAMPER — Feature Ideation

Method Idea Applied
Substitute Replace manual search with Research Agent summaries
Combine Merge booking + itinerary + budget into single "trip cart"
Adapt Shopping cart holds — bundle flight + stay + activities together
Modify Show true total cost (fees + transfers + baggage) upfront
Put to Use Use itinerary during travel: live alerts, rebooking, local tips
Eliminate Remove repeated data entry with persistent traveler profiles
Reverse Start from budget → suggest destinations that fit the number

📁 Project Files

File Description
TripPilot_MultiAgent_Travel_Planning.pptx Full product deck — problem, architecture, ERRC, SCAMPER, KPIs
TripPilot_Pain_Points_Ashok_Ankalla.docx 10 validated user pain points with real consumer sources

📊 Success Metrics

KPI Definition Target
Time-to-Plan Minutes from first input to confirmed itinerary < 15 min (vs 2-4 hours today)
Price Accuracy % match between shown price and final checkout price > 99%
Booking Trust Score User confidence rating pre-checkout (1-5) > 4.2 / 5
Conflict-Free Rate % itineraries with zero time/transfer conflicts > 98%
Agent Alignment Rate % AI recommendations user accepts without change > 75%

💡 Key Learnings

"The market gap isn't information. It's orchestration."

  1. Users don't need more travel data — they need something that connects the data intelligently
  2. The booking step is where trust collapses — human control gates are non-negotiable
  3. "True total cost" is a trust feature, not a pricing feature — it's about honesty, not math
  4. Google shelving its planner validated the thesis: this problem is hard enough to be worth solving

Built as part of BITSoM Executive Program — Product Management with Generative & Agentic AI · 2026 Author: Ashok Ankalla — Enterprise Data & AI Transformation Leader

About

6-agent AI travel planner — Research, Budget, Booking, Itinerary, Context & Monitor agents with human control gates

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors