Map
A geographic view that brings attractions, points of interest and practical park context into one decision surface.
Independent product · Live in production
RideAdviser
A theme-park planning product that brings map context, schedules and personal decisions into one experience — with AI used where it makes the product genuinely more useful.
RRideAdviser
Find attractions, shows and more
25min10min15minNEXT UP · 11:30Adventure route3 planned stops · 1.8 km
RideAdviser · Park map and planning experience
The problem
Better decisions in a changing environment.
Theme-park visits are full of small, time-sensitive choices: where to go next, whether a queue is worth it, which show still fits the day and how to adapt when priorities change. Relevant information is often scattered across a map, a timetable, official pages and a visitor’s own memory.
RideAdviser explores a more useful alternative: one product that can bring together park context, personal preferences and planning signals without pretending that uncertain data is more precise than it is.
The product
Four connected product areas.
The public application brings Map, Shows, Planner and Profile together as the core of the visitor experience.
Shows
Time-sensitive show information that can be considered alongside the rest of a visitor’s day, rather than in isolation.
Planner
A personal itinerary that accounts for priorities, time windows, show choices and the cost of moving through a park.
Profile
A home for saved preferences, plans and account-related product experiences that make the product more personal over time.
RRideAdviser
Find attractions, shows and more
25min10min15minNEXT UP · 11:30Adventure route3 planned stops · 1.8 km
RideAdviser · Park map and planning experience
01Map
02Shows
03Planner
04Profile
AI at work
Automation tied to real product tasks.
Each AI capability starts with a practical goal: reduce repetitive content work or help a visitor make a better decision. Clear inputs, constraints and fallbacks keep the result useful.
01
Content operations
Attraction enrichment with traceable evidence.
A park catalogue rarely arrives complete. RideAdviser can turn an incomplete attraction record into structured editorial data — such as a description, classification, height rules, ride duration, tags and suitability signals.
- Official park domains are the preferred evidence source, with matching alternatives used only when needed.
- The result is constrained to a structured schema and carries confidence, source URLs and a reason when a field cannot be resolved.
- Batching, retry rules, field-level status and execution budgets make this a controllable content workflow rather than an opaque bulk prompt.
02
Planning engine
AI refines a plan; deterministic logic keeps it sound.
The planner starts from a calculated baseline. It combines opening hours, attractions, visitor priorities, height eligibility, selected shows, walking time and available live or historical waiting-time signals.
- An LLM can improve sequencing and explain the result, but its output is schema-validated before it is accepted.
- If an AI suggestion reduces priority coverage, available experiences or route quality, the stronger deterministic baseline is kept.
- The final pass enforces practical rules — including show schedules, route-aware timing and family constraints — after AI refinement, not before it.
03
Visitor assistant
A chatbot that can ask the product for facts.
The assistant is designed around the visitor’s current park and planner context instead of answering from a generic prompt alone. It can request the information it needs through narrowly defined product tools.
- Tools provide relevant weather, archived wait statistics and saved planner state when the visitor’s session permits it.
- Tool calls, message size and follow-up rounds are bounded; unavailable data produces a clear fallback instead of fabricated certainty.
- Personal planning data is only available through an authenticated request, keeping the assistant’s context aligned with product permissions.
04
Visual content
Image generation that respects the real attraction.
Generated imagery can help create consistent visual coverage for a catalogue, but only if it does not redesign the subject. RideAdviser uses official attraction-page images as visual references where they are available.
- Prompts preserve the visible ride system, structure, geometry and permanent scenery while allowing a coherent illustrative treatment.
- Specific policies prevent invented water features or a different ride type when the reference material does not support them.
- A separate visual quality check can reject an image when the result does not match the attraction’s factual identity.
My role
Product thinking through production ownership.
RideAdviser is the clearest public example of independent, end-to-end product work: shaping the concept, making technical decisions, building the experience and owning what happens after the first release.
- Product concept and experience design
- Architecture and technical decision-making
- Full-stack implementation
- AI-assisted feature delivery and validation
- External data and platform integrations
- Deployment and ongoing production ownership
Engineering focus
The hard part is making the parts work together.
The value comes from the connection between park data, visitor needs, route logic, schedules and product operations — not from any individual feature in isolation.
01Changing external data
02Route-aware planning
03Shows and hard time windows
04Personalisation and family rules
05Grounded assistant interactions
06Reliable content automation
What it demonstrates
Independent product capability, made visible.
RideAdviser is evidence of the kind of work I enjoy: taking an ambiguous product problem, choosing practical technology and shipping a coherent, maintainable experience.
- Owning a live product from idea to operation
- Designing AI capabilities around real user and editorial workflows
- Combining deterministic logic with generative models responsibly
- Balancing product UX, backend services and data integrations
- Building safeguards that keep AI output reviewable and useful
Start a conversation
Have a product, system or stubborn software problem?
Tell me what you are trying to build or improve. I'll tell you where I can help.
Discuss a projectBased in Budapest · available across Europe