Assess

Assess by Clustral AI understands what you need, finds the right match, and uses Perfect Corp to show you what it looks like on you.

Comment

Assess — Find the right thing. Then see yourself in it.

Most virtual try-on experiences start with a product.

Assess starts with the person.

We built Assess by Clustral AI for the DevNetwork [API + Cloud + AI] Hackathon 2026 and the Perfect Corp challenge.

The idea was simple: before asking someone to virtually try on something, first understand what they are actually trying to accomplish.

A user can say:

“I need something for a semi-formal dinner, under $150, elegant but not too flashy.”

Assess turns that natural-language request into structured intent — occasion, budget, style, formality, category and preferences — then ranks products against that intent.

The result is three recommendations, each with an explanation of why it fits.

Only after the user selects a product does the visual layer begin.

Clustral finds the match. Perfect Corp shows what it looks like on you.

The user uploads a photo only when there is a product worth trying. Assess sends the selected garment and photo to Perfect Corp's AI Clothes API and returns the virtual try-on result alongside the original image.

How we built it

Assess combines:

  • Azure OpenAI for natural-language intent understanding
  • A deterministic parser as an always-on fallback
  • Clustral's recommendation engine for weighted product ranking
  • SerpApi Google Shopping for live product discovery
  • Perfect Corp YouCam AI Clothes for virtual try-on
  • Xano / Azure Table / memory for persistence
  • Next.js for the guided frontend
  • Azure App Service for deployment

We also designed the system around graceful degradation.

If the language model is unavailable, intent parsing continues through deterministic rules. If live discovery is unavailable, Assess falls back to its curated catalog. If Perfect Corp is unavailable, the recommendation remains usable and the UI clearly tells the user that try-on is unavailable.

We wanted the system to degrade honestly rather than fail silently.

What we learned

The interesting part wasn't simply connecting several APIs.

It was deciding when each technology should be used.

AI shouldn't be called just because it is available. A virtual try-on shouldn't happen before there is something worth trying. Live discovery shouldn't replace a reliable fallback.

That led us to a simple architecture:

Understand → Decide → Visualize

Clustral provides the intelligence that narrows the choice.

Perfect Corp provides the visualization that makes the choice tangible.

Assess connects the two.

The goal isn't to replace the shopper's judgment.

It's to make that judgment easier.

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Assess

Built With

  • ai
  • azure
  • clustral
  • computer
  • corp
  • generative
  • microsoft
  • next.js
  • openai
  • perfect
  • recommendation
  • serpapi
  • systems
  • try-on
  • typescript
  • virtual
  • vision
  • xano
Share this project:

Updates