Inspiration

As head of growth at oro, one of our biggest problems is not being able to outmog our opps.

What it does

  • Upload your opp's fit
  • Finds out where that fit is from
  • Suggests ways to MOG your opps

How we built it

  • Frontend: Next.js App Router pages (/ landing and /stalk analyzer), custom dark UI, card-based results. Image analysis: Gemini multimodal prompts to extract visible clothing items and attributes (color, style, material, fit).
  • Shopping retrieval: Serper shopping API for real product links/images.
  • Fit upgrade flow: Gemini proposes a new outfit direction, then Serper resolves those ideas into real products.
  • Explainability layer: Gemini generates a short rationale between “clothes found” and “your mog fit.” Response shaping: strict JSON parsing, sanitization, and fallback model logic for robustness.

Challenges we ran into

  • Shopping auth/provider mismatch: early 401s required better diagnostics and clearer provider-specific setup.
  • Hallucinated product assets: Gemini-generated links/images weren’t consistently real, so we switched to retrieval-backed product resolution via Serper.
  • “New fit” quality: initial outputs repeated similar garments, so we tightened prompting and similarity filtering to force more distinct outfit families.
  • Tone consistency: balancing meme/brainrot copy with still-usable UX took iteration.

Accomplishments that we're proud of

  • Built a full end-to-end pipeline from image input to shoppable output.
  • Added a fun, distinctive brand voice that matches the “dress better than your opp” concept.
  • Improved reliability by replacing AI-only product URLs with search-backed real listings.
  • Added explainability (why this fit mogs your opp) to make recommendations feel intentional.
  • Shipped fast iteration loops on copy, UX, and ranking behavior based on live testing feedback.

What we learned

  • AI-only generation is great for creative direction, but retrieval is essential for trustworthy commerce results.
  • Prompting alone isn’t enough; adding lightweight post-filters dramatically improves recommendation quality.
  • Error observability (status/body snippets/context) is critical when chaining multiple APIs.
  • Users care as much about tone and vibe as raw functionality for consumer-facing AI products.
  • Small UX wording changes can significantly improve perceived product personality and engagement.

What's next for slay stalker

  • integrate this into oro?

Built With

Share this project:

Updates