Forem: Anfath Hifans

Ceylon Journey Partner - Real-Time AI Travel Partner for Sri Lanka

Anfath Hifans — Mon, 01 Sep 2025 01:44:20 +0000

This is a submission for the AI Agents Challenge powered by n8n and Bright Data

What I Built

I built Ceylon Journey Partner – Real-Time AI Travel for Sri Lanka, an AI-powered travel planning workflow that creates a personalized itinerary for tourists visiting Sri Lanka.

The workflow solves the common problem of overwhelming travel research by automating the process of collecting hotel options from multiple platforms (Booking.com, Airbnb, Agoda) and combining them with user preferences such as travel dates, budget per day, destinations, and activities.

It ensures travelers get a balanced plan that maximizes good weather, minimizes travel time, fits within budget, and matches personal interests.

Preview Workflow

Demo

n8n Workflow

GitHub Gist - Ceylon Journey Partner

Technical Implementation

The workflow is implemented in n8n with the following components:

System Instructions (AI Agent Prompt):

You are a Sri Lanka travel planning expert.  
Using the provided hotel data (Booking.com, Airbnb, Agoda), travel country, destination cities, travel dates, budget per day, traveler count (adults, children, infants, pets), and interested activities:  

Return a complete travel plan that:  
- Covers ALL provided cities between the given dates.  
- Suggests daily activities near each city based on the interests provided.  
- Recommends at least one accommodation option in each city within the budget.  
- Estimates daily costs (accommodation, activities, transport).  
- Highlights potential risks (weather, closures, peak crowd times).  
- Suggests 1–2 backup options if risks exist.

Model Choice: Google Gemini 2.5 Flash (via n8n AI Agent node)
Memory: Stateless (each itinerary generated fresh from real-time inputs)
Tools Used:

Webhook Trigger
- Entry point for user input.
- Captures country, destination cities, travel dates, budget per day, traveler count (adults, children, infants, pets), and interests.
Collect Inputs
- Normalizes the JSON payload for downstream processing.
Hotel Data Scraping with Bright Data
- Three Bright Data Verified Nodes + custom Code nodes filter hotel listings from:
  - Booking.com dataset (gd_m5mbdl081229ln6t4a)
  - Airbnb dataset (gd_ld7ll037kqy322v05)
  - Agoda dataset (gd_m837ssst155rq3a1xo)
- Each platform uses filter queries based on country and destination cities.
- Snapshots are created, checked for readiness, and retrieved.
Snapshot Management
- Snapshot Status node checks if Bright Data snapshot is built.
- If + Wait nodes ensure retry until dataset is ready.
- Switch Node maps dataset IDs to Booking, Airbnb, or Agoda.
Hotel Data Aggregation
- Aggregate nodes collect and unify hotel listings per platform.
- Merge node combines Booking, Airbnb, and Agoda data into a single dataset.
User & Hotel Data Merge
- User preferences and hotel datasets are aggregated into a structured input for the AI agent.
AI Agent Node (Google Gemini 2.5 Flash)
- Prompted as a Sri Lanka travel planning expert.
- Generates a day-by-day itinerary that:
  - Covers all destination cities within travel dates.
  - Suggests activities aligned with user interests.
  - Recommends budget-matching accommodations (Booking.com, Airbnb, Agoda).
  - Estimates daily costs (hotel, transport, activities).
  - Highlights risks (weather, closures, crowds).
  - Suggests backup options.
- Output format: Gmail-compatible HTML (tables, headings, booking links).
Email Delivery
- Final travel plan sent to the user via the Gmail Node.
- HTML itinerary is styled and directly viewable in email.

Bright Data Verified Node

Bright Data’s Verified Node was used to scrape hotel availability, pricing, and location details from Booking.com, Airbnb, and Agoda in real-time.
This ensures itinerary recommendations are always up-to-date with accurate prices and availability.

Journey

I wanted to solve a real-world tourism problem in Sri Lanka — how to plan trips without juggling dozens of tabs.

The first challenge was merging heterogeneous data. I overcame this by using n8n’s Merge Node and custom function nodes to structure everything.
Next, I crafted an AI Agent prompt that balanced flexibility with structure, so itineraries are clear but still personalized.
Finally, integrating Bright Data ensured the agent works with real-world, ever-changing data rather than static APIs.

Key challenges I overcame:

Merging different hotel data sources into a single dataset
Normalizing budget and traveler preferences into a scalable filter
Generating itineraries that balance weather, budget, and travel time
Automating the entire process into a single n8n workflow

Through this, I learned how powerful Bright Data + n8n can be when combined with AI — enabling real-time decision-making for travelers.

🚀 With Ceylon Journey Partner, anyone can instantly generate a real-time, AI-powered Sri Lanka itinerary — combining the magic of travel with the reliability of structured data.

Future Extention

💡 This is just the start — I'm exploring ways to extend it with:

Future Features

Local Accommodation 🏨: Hotels, 🏝️Resorts, 🏖️Beach Bungalows, 🏡Villas, ⛺Glamping Sites, 🏕️Camping Sites, and 🏠Homestays.
Real-time Weather Data ☀️: Providing up-to-the-minute weather forecasts for destinations.
Event Suggestions 🎶🎊: Curated lists of local events and festivals.
Travel Optimization ✈️: Intelligent routing for flights, trains 🚆, buses 🚌, and taxis 🚕.

🍽️ NutriGuard AI: An Intelligent Meal Planner with Toxic Food Combo Alerts and Condition-Based Food Warnings

Anfath Hifans — Mon, 07 Jul 2025 04:12:41 +0000

This is a submission for the Runner H "AI Agent Prompting" Challenge

🧠 Concept Overview

A multi-agent AI system that:

Plans meals based on dietary preferences, goals, and schedule.
Flags harmful food combinations (e.g., dairy + citrus, iron + calcium blockers).
Warns against foods based on medical conditions (e.g., “avoid grapefruit if on statins” or “no nightshades for arthritis”).

✅ Why It’s Unique & Useful

Combines meal planning, dietary safety, and personalized health guidance.
Solves a real-world daily problem — what to eat safely, deliciously, and conveniently.
Useful for people with diabetes, hypertension, thyroid issues, heart-issues, PCOS, alergic, or on specific meds.
Stands out by not only recommending what to eat, but also what not to combine or consume.

What I Built

NutriGuard AI is a prompt-engineered multi-agent workflow designed to provide personalized meal planning, toxic food combination detection, and health-condition-aware dietary filtering.

This Runner-H AI system solves a common yet critical problem:

“What should I eat that’s safe, healthy, and suited to my condition?”

NutriGuard allows users to input dietary preferences (e.g., vegetarian, keto), medical conditions (e.g., heart-issues, thyroid, PCOS, diabetes), and dislikes/allergies. The system:

Creates a smart daily or weekly meal plan
Warns about harmful food pairings
Flags ingredients that may worsen the user's health condition
Generates a shopping list with safe substitutions

Each function is powered by dedicated prompt-driven agents, making the system modular, transparent, and adaptable.

Demo

Example User Flow
User Input:

“I’m vegetarian with thyroid issues. I don’t like mushrooms. Plan my meals for the next 3 days, and avoid foods that interfere with my medication.”

Output:

Meal Plan:
- ✅ Breakfast: Oatmeal with almonds and berries (no soy or broccoli)
- ⚠️ Lunch: Lentil salad with lemon — replaced spinach (high in oxalates)
- ✅ Dinner: Paneer tikka with quinoa and turmeric vegetables1. Actual numbers don't matter, just that it's a number
🛑 Warning: Avoid soy, cabbage, and cauliflower — they can suppress thyroid function.
🛒 Shopping List: Includes safe alternatives and quantities.

Screenshot

Link
https://runner.hcompany.ai/chat/4819c15f-43ca-4f18-83a9-63855ba4539d/share

How I Used Runner H

I used Runner H to build a multi-step, prompt-based automation where each stage of the meal planning process is handled by a dedicated AI agent:

Agents:

🧠 Meal Planner Agent – Generates meals based on user preferences, schedule, and dietary needs.
⚠️ Toxic Combo Checker – Detects food pairings that may be harmful (e.g., milk + citrus).
🚫 Condition Filter Agent – Flags and substitutes foods harmful to medical conditions.
🛒 Shopping List Agent – Compiles a clean, ingredient list optimized for nutrition and safety.
📚 Nutrition Explainer (Optional) – Explains flagged items and nutritional choices.

Workflow in Runner H:

Each step is chained using clear prompt-based tasks.
User input is parsed once, then passed through each agent.
Agents respond with structured JSON blocks for easy integration with UI or PDF generation.

Sample Prompt for the Toxic Combo Agent

You are a food safety expert. Analyze the above-provided meal and flag
any known harmful food combinations. Use evidence-based knowledge 
(e.g., high-iron foods mixed with calcium inhibit absorption, 
citrus + milk may cause digestion issues, etc.).

Sample Prompt for Condition Filter Agent

You are a clinical nutritionist. Based on the condition: “Hypothyroidism”,
scan the above-provided ingredients and remove or replace any foods
that negatively affect this condition or common medications related
to it. Explain your choices briefly.

This modular approach allows other users to replicate, extend, or plug in their own use cases (e.g., low-FODMAP diet, PCOS-safe foods).

Use Case & Impact

NutriGuard AI is built for:

Individuals with chronic conditions (diabetes, thyroid issues, hypertension, PCOS)
Health-conscious eaters seeking personalized meal planning
Caregivers or nutritionists creating plans for others
Anyone overwhelmed by conflicting online food advice

Real-World Impact:

Reduces the risk of harmful food interactions
Ensures condition-specific dietary compliance
Makes meal planning easy and safe
Saves time and avoids misinformation

It’s like having a personal dietitian + food scientist + chef all in one AI flow.

Social Love

Sharing this on X/Twitter and LinkedIn