# We Chose Human Intelligence Over AI — And It Worked Better Than We Hoped

At a recent *AI for Good* Hackathon, our team, *Rooted in Resilience*, set out to solve a real-world challenge brought to us by **Refugee & Immigrant Transitions (RIT)** — a nonprofit that supports immigrants and refugees through English language and job-readiness programs.

Their process for collecting feedback was painfully analog:

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1749057050515/aa40e404-2188-4486-a05f-499c69ec21a2.jpeg align="center")

> Distribute printed surveys.  
> Students handwrite their responses.  
> Volunteers manually enter data into spreadsheets.  
> Occasionally, mistakes creep in.  
> Always, it takes too much time.

So we asked the obvious question: *Can AI help?*

## Why AI *Wasn’t* the Best Answer

Like most teams at the hackathon, we started with modern LLMs — GPT-4, Claude, Gemini — to interpret scanned forms. The results?

* Text fields: surprisingly accurate.
    
* Checkboxes: 40–60% failure rate, even with top-tier models.
    
* Cost: way too high for a nonprofit running on limited funds.
    

We realized quickly that no matter how sophisticated the AI, checkbox detection was too flaky — and definitely not budget-friendly.

---

## So We Did the Unthinkable:

### We Chose to Be Intelligent Instead

Instead of brute-forcing AI into a problem it wasn’t designed to solve, we **designed around it**.

We built **SnapScan** — a complete survey automation system that works *without* needing advanced AI for every input. Here's how it works:

---

## SnapScan: Our Solution

### 1\. **Custom Form Generator**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1749056891328/08309103-2e58-4eb6-ad7b-91692b5c6b5a.png align="center")

We built a tool that lets volunteers recreate RIT’s paper forms — preserving the layout and structure, but also adding two major upgrades:

* **Exact box coordinate mapping** (crucial for visual detection)
    
* **A unique QR code** to tag each form to the right spreadsheet
    

### 2\. **Scan and Detect**

Once forms are filled and scanned:

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1749056942569/34f95da4-542c-4227-ba33-e795cc66c23d.jpeg align="center")

* The QR code tells the system which questions and spreadsheet to sync with
    
* OpenCV checks for checkbox marks using pixel-perfect box mapping
    
* OCR handles the occasional handwritten free-text fields
    

### 3\. **Auto-Sync to Google Sheets**

Every response is instantly logged into a shared spreadsheet — no sorting, no manual entry, no fuss.

---

## A Better Demo Than AI

Here’s how we introduced it at the hackathon:

> *"This is the form RIT currently uses — built in Word. Here’s what it looks like in our system. Same format, but now there’s a QR code."*  
> *"Alberto — our fictional RIT volunteer — prints and distributes the forms. Once they’re filled out, he uploads all scanned forms to a Google Drive folder."*  
> *"SnapScan reads the QR code, detects the checkboxes, extracts the handwriting, and updates the spreadsheet. All automatically."*

%[https://www.youtube.com/watch?v=8ZoVevdKZik] 

---

## Why It Worked

* **100% checkbox accuracy** thanks to custom layout and OpenCV
    
* **Zero per-form AI costs** — a huge win for nonprofits
    
* **Fully automated pipeline** — from scan to sheet
    
* **Volunteer-friendly** — no training needed, no tech knowledge required
    

---

We didn’t win the hackathon.  
We didn’t use the flashiest LLM stack.  
We didn’t even qualify under the "AI requirement" strictly.

But we built something that **actually works** — accurately, affordably, and at scale — for a nonprofit that truly needed it.

And sometimes, that’s worth more than a trophy.

---

### The Team:

1. Sanath Swaroop Mulky
    
2. Aditi Dani
    
3. Manav Chandani
    
4. Prithvi Elancherran
    
5. Armin Foroughi
    

---

### The Tech Stack:

* OpenCV (checkbox detection)
    
* ***pytesseract*** OCR(text extraction)
    
* React + Firebase (frontend & auth)
    
* Google Drive + Sheets API (storage & sync)
    
* Node.js (backend)
    

---

### Want to Try SnapScan or Contribute?

We're exploring open-sourcing the tool for other nonprofits.  
Leave a comment or DM if you're interested in collaborating or piloting it with your organization.