alpAI

Applied AI
Product Case Study

Building an applied AI system that connects mountain routes, conditions, weather, safety and artificial intelligence.

Explore the productHow the AI works ↓
  • AI
  • Product Engineering
  • Mobile
  • Maps
  • Safety
  • Structured Data

01 — Problem

Problem

Planning a mountain trip means connecting information from multiple sources: routes, difficulty, elevation gain, weather, conditions, huts, access, equipment and safety guidance.

alpAI explores how structured data, conventional application logic and generative AI can bring this information together as clear, useful support for trip planning.

AI is not the source of facts about mountain routes. Its role is to interpret, personalise and explain data.

02 — Product

Product demo

Screens from the actual product, also used on the existing website.

alpAI home screen with routes, weather and an AI insight
Home / mountain overview
Detailed mountain weather view with warnings
Weather / conditions / warnings

03 — Product engineering

My role

The project has been developed as a real product, not a demonstration chatbot. The work spans product decisions, structuring domain knowledge, maps and data, AI integration, and iterative improvements to the user experience.

Product concept
AI workflows
Data architecture
Mobile application
UX/UI iteration
API & data integration
Testing & validation

04 — Architecture

How alpAI works

01

User

  • User preferences
  • Experience / goals / selected trip
02

Structured data

  • Routes
  • Trailheads
  • Elevation
  • Difficulty
  • Duration
03

Context

  • Weather
  • Mountain conditions
  • Huts
  • Transport
  • Daylight
04

Rules & validation

  • Hard constraints
  • Safety checks
  • Data validation
05

AI layer

  • Interpretation
  • Personalisation
  • Explanation
06

User output

  • Route recommendation
  • Warnings
  • Alternatives
  • Contextual explanation

LLM is a reasoning and communication layer — not the source of geographical facts.

05 — Example workflow

From request to recommendation

User input
  • Saturday
  • 5–6 hours
  • Good fitness
  • I want a more demanding marked route
alpAI

Routes · Weather · Conditions · Safety · User context

Recommendation

Viševnik from Rudno polje

✓ suitable duration

✓ conditions are suitable

✓ finish before sunset

⚠ it may be colder above 2,000 m

⚠ check the required equipment

Process illustration — the data is not presented as a current real-time forecast.

06 — Grounding & validation

Building reliable AI

Development exposed an important problem: a generative model can provide geographical information that sounds entirely plausible but is wrong. Queries about Triglav could produce an unsuitable trailhead. This led to an important architectural decision.

Before

User

↓

LLM

↓

Route information

Problem: plausible ≠ correct
After

Verified / curated route data

↓

Application logic

↓

AI interpretation

↓

User

Use AI to interpret facts. Do not use AI to invent facts.
  • Grounding
  • Hallucination awareness
  • Deterministic + generative logic
  • Validation
  • Safety-critical thinking

07 — Applied AI

Three core AI use cases

01

AI Route Recommendation

  • čas / time
  • izkušnje / experience
  • kondicija / fitness
  • želje / preferences
  • razmere / conditions
02

AI Route Check

  • vreme / weather
  • čas / time
  • zahtevnost / difficulty
  • logistika / logistics
  • opozorila / warnings
03

AI Mountain Assistant

  • razlaga razmer / conditions
  • priprava opreme / equipment
  • orientacija / orientation
  • varnost / safety
  • informacije o turi / trip context

The AI assistant does not replace a mountain guide, mountain rescue services, official weather sources or human judgement.

08 — Safety

Safety by design

Verified data where possible
Clear uncertainty
Hard safety rules
Context before recommendation
Emergency information remains deterministic
AI does not replace official information
STOPSTOP / Orientation
112SOS / 112
◌Conditions
!Warnings

09 — Technology

Tech stack

Mobile

Flutter

Cross-platform mobile product

Frontend

Next.js + React

Public website and case study

Backend

Node.js + Express

AI and transport service endpoints

AI

OpenAI API

Interpretation and conversational layer

Maps

Mapbox + OpenStreetMap

3D maps, routes and geographic data

Weather / data

ARSO + Meteoblue + Open-Meteo

Weather context and forecasts

Services

Firebase + Sentry

Authentication, data and diagnostics

Deployment

Vercel

Website deployment configuration

10 — Process

Building the product

  1. 01Problem definition
  2. 02UX and information architecture
  3. 03Route/data modelling
  4. 04First working product
  5. 05AI integration
  6. 06Testing real-world edge cases
  7. 07Improving reliability
  8. 08Current development

11 — Reflection

What I learned

AI ≠ database

Generative models are excellent at interpretation and communication, but they are not a reliable primary source of domain facts.

AI UX matters

Returning an answer is not enough. Users need to understand why something is recommended and what deserves their attention.

Safety changes architecture

In a mountain app, a wrong answer is more than poor UX. Some rules and data therefore need to remain deterministic.

Real products expose edge cases

A real application quickly exposes what chatbot prototypes often hide: data, state, errors, user context, mobile UX and integrations.

alpAI

alpAI is an evolving product.

It also serves as a practical environment for exploring how useful, reliable AI can be integrated into a real-world application.

Explore the productContact