How I helped shift prototyping at Stepstone from static simulations to functional, data-driven products, using Figma Make, AI models, and real data integrations to test real behaviour instead of imagined scenarios.
Traditional design prototypes simulate interactions, but they don't behave like real products. During research, this creates a fundamental disconnect, users are asked to pretend, inputs don't behave realistically, data is static or fabricated, and insights are limited by the prototype itself.
As our ambition for research and product quality increased at Stepstone, traditional tools created hard constraints: no real data, no logic, no persistence, no adaptability.
We weren't testing real behaviour. We were testing how well users could imagine.
Advancements in AI and tools like Figma Make introduced a new possibility, prototypes that could process real user input, connect to live datasets, adapt dynamically to user behaviour, and function similarly to production systems.
Prototypes that behave like products.
I focused on combining three elements to make this possible:
Using structured datasets, real job listings, real content, prototypes could reflect actual product scenarios rather than fabricated ones.
Natural language input could be interpreted and transformed into structured queries, enabling intelligent, adaptive behaviour without predefined filters.
User interactions stored and reused across sessions via Supabase, creating continuity that mirrors real product behaviour.
Logic that responds to what the data returns, broadening or refining results dynamically based on volume and relevance.
Instead of traditional filters, users describe what they're looking for in natural language, and the system interprets intent rather than waiting for structured input.
A user inputs a query like "UX Designer in London for at least £50k". AI interprets job title, location, and salary expectations, maps them to structured fields, and returns results dynamically from real data.
Search results can overwhelm or underdeliver, too many results with no guidance, or too few with no clear path forward.
Adaptive logic: if results < 50, broaden the search. If results > 50, refine it. A balanced experience that mimics intelligent system behaviour without manual filtering.
Research participants are often asked to engage with artificial scenarios, pretending data is real, imagining how a feature might work.
Prototypes designed to accept real user inputs, use real-world data, and respond dynamically, allowing participants to engage with scenarios relevant to their actual lives.
Traditional prototypes reset after each interaction, no memory, no continuity, no way to test multi-step or returning-user journeys.
Using Supabase to store user sessions, inputs persisted across flows, prototypes could "remember" users and simulate real product behaviour across multiple interactions.
This work redefined what a prototype could be, not just in terms of fidelity, but in terms of capability and purpose.
Prototypes became environments for testing real behaviour, not imagined scenarios.