Easify

Find the right sponsor-company fit in minutes.

Comment

Inspiration

Sponsorship outreach usually starts with a lot of manual work: searching company websites, reading scattered pages, guessing whether a partnership will fit, and then writing outreach emails from scratch.
I wanted to build something that turns this process into a faster, data-driven workflow, especially for students, organizers, and early teams who do not have a dedicated partnerships team.

That idea became Easify: an AI-powered sponsorship intelligence platform that helps users quickly understand company-program fit and generate a tailored sponsorship letter.

What It Does

Easify takes two inputs:

  1. A sponsorship program profile (name, type, goals, target audience, ask)
  2. A target company website URL

It then:

  • Extracts surface-level website signals
  • Analyzes alignment between the company and the program
  • Returns structured match insights
  • Generates a ready-to-send sponsorship letter
  • Shows token usage and estimated cost transparency

How I Built It

I built Easify as a full-stack app:

  • Frontend:React + TypeScript + Vite
  • Backend: Node.js + Express + TypeScript
  • Extraction layer:** Axios + Cheerio
  • AI layer: Gemini (primary), with optional multi-provider-ready architecture
  • Deployment: Vercel

Core flow

  1. User saves program details once
  2. User enters company URL
  3. Backend extracts key website sections (mission, projects, methods, outreach, etc.)
  4. AI generates:
    • company profile summary
    • sponsorship overlap analysis
    • personalized letter
  5. Frontend shows results + token usage breakdown

I also added product polish:

  • provider status checks via /api/health
  • retry/fallback behavior for unstable network/provider responses
  • a cleaner iterative UI for checking multiple companies without re-entering program info
  • visual branding updates (logo + improved deck/screenshots)

Challenges I Faced

1) Noisy web data

Websites are inconsistent and messy. Extracting useful text without pulling too much irrelevant content was difficult.
I solved this by limiting crawl depth and using targeted keyword-based section extraction.

2) Cost and token limits

Early prompt flows were too expensive for free-tier usage.
I reduced token usage by:

  • compacting extracted input
  • reducing output limits
  • combining expensive analysis steps
  • enforcing concise JSON output formats

3) Reliability in deployment

Cross-origin issues, endpoint failover, and intermittent backend connectivity created real-world friction.
I fixed these by tightening CORS, improving fallback API resolution, and adding clearer frontend health/error states.

4) AI output quality

Generic responses were not useful enough for real sponsorship outreach.
I improved this by making prompts stricter and evidence-focused so outputs stayed specific and actionable.

What I Learned

This project taught me that building with AI is not just “call model, get answer.”
The real work is in:

  • data shaping
  • prompt engineering
  • reliability under deployment constraints
  • user experience for iterative workflows
  • cost-performance tradeoffs

I also learned how important observability is. Even a simple token display helps users trust and control AI usage.

A Small Formula I Used

I used a simple token-cost estimate in the app:

$$ \text{Estimated Cost} \approx \text{Total Tokens} \times 0.000003 $$

This keeps AI usage transparent for users testing under limited budgets.

What’s Next

Planned next steps:

  • CRM/email integrations
  • better multi-company comparison views
  • improved sponsor recommendation scoring
  • proposal/PDF generation pipeline

Built With

Share this project:

Updates