PathFinders

An AI-powered school advisory system for Singapore's Secondary 1 Posting process

About

PathFinders is an AI-powered school advisory system designed to support parents in navigating Singapore's Secondary 1 Posting process with clarity, accuracy, and personal relevance.

Unlike today's fragmented landscape where parents rely on SchoolFinder, scattered school websites, SchLah, and informal word-of-mouth, PathFinders introduces the first unified, rule-driven AI engine aligned strictly with Ministry of Education (MOE) guidelines.

The system consolidates official cut-off point (COP) data, programme information, and behavioural insights into a single platform eliminating guesswork and significantly reducing the time parents spend researching schools.

Inspiration

The idea for PathFinders came from a deeply personal place. Years ago, I didn't get into the secondary school I wanted not because I wasn't qualified, but because my family simply wasn't well-informed about how the school system worked. We didn't know about programme differences, or even which schools might have been a better fit for my interests and learning style. We were navigating in the dark, relying on fragmented information and word-of-mouth advice that often turned out to be outdated or inaccurate.

Recently, I watched my younger cousin go through the exact same struggle during their S1 Posting process. Their family was overwhelmed—juggling information from MOE's website, SchoolFinder, SchLah, and dozens of individual school websites, trying to piece together a picture of which schools were realistic options and which would truly suit my cousin's strengths and interests. I saw the stress, the uncertainty, and the fear of making the "wrong" choice. And I realized: this shouldn't still be happening in 2025.

Every year, many families in Singapore face this same daunting task. Many, like mine, come from backgrounds where navigating the education system doesn't come naturally where parents don't have insider networks or prior experience with the process. The information gap isn't just inconvenient; it's inequitable. Families with resources and connections get better outcomes, while others miss opportunities simply because they didn't know better.

I didn't want my cousin or any other family to experience what we went through. That's why we built PathFinders: to level the playing field and ensure every family, regardless of background or resources, has access to the same quality of guidance when making one of the most important educational decisions for their child.

PathFinders is about turning my family's struggle into a solution that empowers thousands of others because no child should miss out on the right school simply because their family didn't have the right information.

Features

Core Capabilities

AI-Powered Chatbot - Conversational interface for school queries, recommendations, and comparisons
Student Profile Quiz - 12-question assessment to understand child's interests, learning style, and personality
Personalised Recommendations - Curated list of 5 suitable schools based on AL score, priorities, and quiz results
School Comparison - Side-by-side analysis of up to 3 schools with detailed attributes
Appeal Workflow - Guided process for eligible school appeals with draft email generation
Shortlist Management - Save and track schools of interest

Data Integration

140+ Secondary Schools - Comprehensive database with COP ranges, programmes, and CCAs
Real-time Eligibility - Automatic filtering based on student's AL score
Programme Information - IP, SAP, ALP, LLP, EAP, and autonomous school details
CCA Database - Complete co-curricular activities for all schools
Transport Routes - Public transport directions via OneMap API integration

Safety & Compliance

MOE-Aligned Responses - All recommendations follow official guidelines
Anti-Hallucination Layer - Prevents fabrication of school information
Factual Attributes Only - Uses structural school data, not reputational claims
Official Terminology - Consistent use of MOE-approved language

How We Built It

Platform

PathFinders was built entirely on Lovable, an AI-powered development platform that enables rapid prototyping and deployment of web applications. Lovable's integration with Supabase provided the backend infrastructure for:

Edge Functions (AI chatbot logic)
Authentication system
Secrets management for API keys

Technology Stack

Layer	Technology
Frontend	React, TypeScript, Vite
Styling	Tailwind CSS, shadcn/ui
Backend	Supabase Edge Functions
AI Model	Claude (Anthropic) via API
Routing	React Router
State Management	React Hooks, Session Storage
Email	EmailJS
Maps	OneMap API

Architecture

The system employs a modular agent architecture:

Profiling Agent - Collects and processes student information via quiz
Attribute Encoder - Maps quiz responses to school preference weights
Matching Engine - Scores schools based on eligibility and fit
Compliance Validator - Ensures recommendations are MOE-aligned
Response Generator - Formats output with appropriate disclaimers

┌─────────────────────────────────────────────────────────────┐
│                      User Interface                         │
│  (Chat Interface, Quiz Forms, Comparison Tables)            │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                    Edge Function Layer                      │
│  (school-chatbot, onemap-route)                             │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                     AI Processing                           │
│  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐       │
│  │Profiling │ │Attribute │ │Matching  │ │Compliance│       │
│  │Agent     │ │Encoder   │ │Engine    │ │Validator │       │
│  └──────────┘ └──────────┘ └──────────┘ └──────────┘       │
└─────────────────────────────────────────────────────────────┘
                              │
                              ▼
┌─────────────────────────────────────────────────────────────┐
│                     Data Layer                              │
│  (School Database, COP Data, CCAs, General Info)            │
└─────────────────────────────────────────────────────────────┘

Challenges Faced

1. Data Collection & Processing

Challenge: Gathering comprehensive, accurate school data from multiple sources.

Solution: We consolidated data from:

MOE official cut-off points (140+ schools with PG1, PG2, PG3 ranges)
SchLah for additional school attributes
MOE CSV exports for CCAs and general information
Manual verification for programme details (IP, SAP, ALP, LLP)

The data required extensive cleaning and standardisation to ensure consistency across sources.

2. AI Recommendation Justification

Challenge: Ensuring AI recommendations are explainable and justified, not just "black box" outputs.

Solution: We implemented:

Transparent scoring based on explicit criteria (AL eligibility, priority weights, quiz alignment)
Structured reasoning where the AI explains why each school matches the student's profile
Factual grounding using only verifiable school attributes
Anti-hallucination safeguards that prevent the AI from inventing school features or making reputational claims

3. Balancing Personalisation with Accuracy

Challenge: Providing personalised recommendations while staying strictly within MOE guidelines.

Solution: The system uses a layered approach:

Hard filters - AL score eligibility (non-negotiable)
Soft scoring - Priority weights and quiz alignment
Compliance layer - Validates all outputs against MOE rules

4. User Experience for Non-Technical Users

Challenge: Creating an interface accessible to parents who may not be tech-savvy.

Solution:

Conversational chat interface (natural language interaction)
Visual priority sliders instead of numerical inputs
Step-by-step guided workflows
Mobile-responsive design

What We Learned

Domain Expertise Matters - Understanding the S1 Posting process deeply was essential for building a useful tool
Data Quality is Foundational - The AI is only as good as the data it operates on
Safety by Design - Building guardrails from the start is easier than adding them later
User-Centric Development - Continuous feedback from parents helped prioritise features
AI + Rules = Trust - Combining AI flexibility with rule-based compliance builds user confidence

🏃 Getting Started

Prerequisites

Node.js 18+ and npm
Supabase account (for backend services)

Installation

# Clone the repository
git clone https://github.com/mcspicyupsiz/schoolpath-finder.git

# Navigate to project directory
cd pathfinders

# Install dependencies
npm install

# Start development server
npm run dev

Environment Variables

The following environment variables are configured automatically via Lovable Cloud:

VITE_SUPABASE_URL
VITE_SUPABASE_PUBLISHABLE_KEY

Additional secrets (configured in Supabase):

ANTHROPIC_API_KEY - For Claude AI
ONEMAP_TOKEN - For transport routes

Project Structure

src/
├── components/
│   ├── chat/           # Chatbot components
│   ├── landing/        # Homepage sections
│   ├── layout/         # Header, Footer
│   ├── profile/        # Student quiz components
│   ├── recommendations/# School cards, filters
│   ├── school/         # School detail components
│   └── ui/             # shadcn/ui components
├── data/
│   ├── moe-schools-cop.json    # Cut-off point database
│   ├── ccas.csv                # CCA information
│   └── general-school-info.csv # School general info
├── pages/
│   ├── Index.tsx       # Homepage with chatbot
│   ├── Profile.tsx     # Student profile quiz
│   ├── Recommendations.tsx
│   ├── SchoolDetail.tsx
│   └── Shortlist.tsx
└── supabase/
    └── functions/
        ├── school-chatbot/   # AI chatbot logic
        └── onemap-route/     # Transport routing

Sustainability

PathFinders is built to withstand changes in Singapore's evolving education landscape through adaptive design and flexible data architecture.

Resilience to Policy Changes

Singapore's education system undergoes periodic reforms—from PSLE scoring changes (T-score → Achievement Levels) to new school programmes and posting criteria. PathFinders is designed to adapt seamlessly:

Modular Data Architecture

Decoupled Data Layer - School information, COP data, and MOE rules stored separately from application logic
Schema Versioning - Each data structure has version tags, allowing the system to handle multiple formats simultaneously during transition periods
Hot-Swappable Datasets - When MOE announces changes (new AL bands, different COP calculations, revised posting rules), we simply swap out old data files with updated ones—no code rewrite needed

Examples of Adaptability:

2021 PSLE Scoring Reform - When Singapore moved from T-scores to Achievement Levels, a system like PathFinders would only need to update the scoring lookup tables, not the entire recommendation engine
New School Programmes - If MOE introduces new initiatives (like replacing LLP/ALP with different frameworks), we add new attribute fields without breaking existing functionality
COP Calculation Changes - Should MOE modify how cut-off points are determined, the eligibility filter logic can be updated independently

Future-Proof Design

Configurable Rule Engine

MOE guidelines encoded as configuration files, not hardcoded logic
When posting rules change (e.g., new tie-breaker criteria, modified appeal eligibility), we update JSON configurations rather than rebuilding the system
Compliance validator automatically adapts to new rule sets

Extensible School Attributes

Database schema designed to accommodate new school characteristics
Adding new data points (e.g., "sustainability programmes," "mental health resources") requires only dataset updates, not system redesign
Legacy attributes gracefully deprecated without breaking existing functionality

API-First Integration

Ready to consume official MOE APIs if/when they become available
Can integrate real-time data feeds instead of manual updates
Third-party education platforms can be plugged in as they emerge

Annual Maintenance Workflow

MOE releases new COP data (typically November/December)
Data team validates and formats new dataset
Swap data files in production environment
Automated tests verify recommendation accuracy
System live with updated information—zero downtime

Future Enhancements

[ ] Real-time COP updates when MOE releases new data
[ ] School visit booking integration
[ ] Parent community Q&A features
[ ] Multi-language support (Chinese, Malay, Tamil)
[ ] Supporting students through every transition—from Secondary School to Poly/ITE/JC, and onwards to University