Frontkom
Reis Nordland·Artificial Intelligence

How Reis Nordland Cut Customer Inquiries by 55% Using AI

Frontkom developed a scalable, modular AI platform built on IBM technology that delivers an intelligent travel assistant for public transport in Nordland. The result: faster responses, fewer inquiries, and a better travel experience.

Kunde

Reis Nordland (Nordland County Municipality)

Bransje

Public Transport / Public Sector

Oppdrag

Develop an AI-driven travel assistant that gives travellers natural language answers about timetables, delays, and travel information — around the clock, in multiple languages.

Teknologi

LangflowIBM Cloud Databases for PostgreSQLGemini 2.5 FlashKubernetesAWSFastly CDNSupabaseReactVercelEnTur APIGoogle Places

Background

The Challenge: Intelligent Passenger Assistance

Reis Nordland is the regional public transport provider for Nordland county in Norway. The company operates bus and ferry routes across a vast northern territory, managing timetables, ticketing, and travel information for passengers.

Reis Nordland was overwhelmed by repetitive customer service inquiries and needed to provide travel information around the clock. They required a solution capable of delivering precise, context-aware answers — while operators retained full control over what the model responds.

The Solution

An AI Travel Assistant That Understands Natural Language

Frontkom developed and delivered a generative AI travel assistant that lets travellers ask questions in natural language — such as "When does the next ferry from Bodø depart?" or "Is route 300 delayed?" — and receive precise, up-to-date answers through a conversational interface in multiple languages.

The assistant connects to real-time traffic data and structured internal information such as timetables, traffic updates, frequently asked questions, and service policies. The goal was to deliver fact-based, context-anchored answers while operators retain full control.

Results by the Numbers

55%Reduction in customer inquiries after six weeks
1,300+Inquiries resolved in the first week
50%Faster response time at the contact centre
24/7Available in multiple languages

Technical Architecture

Component-Based, Transparent, and Scalable

To support rapid iteration while maintaining control and observability, Frontkom built a component-based architecture centred around IBM technologies and open-source tools.

Frontend and Dashboard

A lightweight, embeddable chat widget built with JavaScript and React integrates directly into Reis Nordland's website and mobile applications. The widget is connected to a secure dashboard built on Supabase, allowing operators to manage AI agents, datasets, and knowledge bases from a single user-friendly console.

Orchestration Layer

The orchestration layer is powered by Langflow, IBM's low-code environment for building AI agent applications. Langflow handles logic, prompt design, and response shaping behind each AI agent. This makes it possible to update components — including language models, data connectors, or workflows — independently of each other without affecting production stability.

Infrastructure

Langflow runs on Frontkom PaaS, a Kubernetes-based infrastructure hosted on AWS. The platform uses Horizontal Pod Autoscaling (HPA) and Pod Disruption Budgets (PDB) to maintain reliability under varying traffic loads. A Fastly CDN ensures fast global delivery for both end users and administrative interfaces.

Data and Storage

All conversation data, agent telemetry, and configuration metadata are stored in IBM Cloud Databases for PostgreSQL. This provides both scalability and governance, with version control tracking every flow and dataset used in production.

Agent Flow

How the AI Agent Works

Within Langflow, each agent follows a defined flow:

  1. Receive input and retrieve current instructions from the database
  2. Select a model via middleware such as OpenRouter or Vercel AI Gateway
  3. Activate tools to retrieve real-time or structured data
  4. Shape and return the response back to the chat interface

Key tools include PlanTrip (EnTur), which combines route, departure, and delay data, and FindPlace/SearchPlace (EnTur + Google Places), which resolves known stops and discovers new places of interest. This architecture ensures the model's responses are both fact-based and visually consistent.

Principles

Lessons Learned from the Project

Building the Reis Nordland assistant highlighted a number of practical and repeatable design principles:

  • Structure drives precision: Strict data schemas and modular flows prevent unpredictable behaviour
  • Iterate continuously: Feedback loops from real usage are essential for improving prompts and outputs
  • Transparency builds trust: Comprehensive observability simplifies debugging and compliance
  • Architecture matters: Component isolation made it possible to update the language model without downtime
Just weeks after launch, the new AI assistant has resolved over 1,300 inquiries. Reis Nordland describes their AI assistant as a 'highly valuable team member'.
Reis NordlandNordland County Municipality

Impact

Reliable AI for the Public Sector

Reis Nordland's assistant now handles thousands of daily interactions, giving travellers fast, multilingual answers. Updating the application is a far simpler process — the team can roll out new features or datasets in hours rather than weeks.

More importantly, the same architecture is being reused for other Frontkom clients — demonstrating that production-grade generative AI is repeatable when the foundation is modular, observable, and governed.

Want to know more about what AI can do for your organisation?

Get in touch for a no-obligation conversation about how we can help you with AI solutions tailored to your needs.

Contact us