ConsilAI

An AI-powered classroom management platform that helps teachers profile students, optimize seating layouts, and generate evidence-backed learning plans — all from one interface.

ConsilAI pairs a Next.js teacher dashboard with a multi-stage AI pipeline: deterministic keyword extraction, automated web research scraping, and LLM-generated intervention plans grounded in real sources. Built as a monorepo with Supabase for auth, persistence, and row-level security.

---

At a Glance (Resume Metrics)

Metric	Value
Total source lines of code	5,663 across 51 source files
Languages	TypeScript, JavaScript/JSX, Python, SQL, CSS
Git commits	61
Contributors	4 developers
Active development period	27 days (Nov 14 – Dec 11, 2025)
Monorepo packages	4 (frontend, backend, ai, scraper)
Frontend pages & routes	10 (9 UI pages + 1 API route)
React components	14 reusable components
Database tables	6 PostgreSQL tables
SQL migrations	4 (530 lines of SQL)
Row-Level Security policies	12 policies across all tables
Stored procedures (RPC)	2 (create_student_occurrence, create_class_occurrence)
SPED issue categories recognized	12 (dyslexia, ADHD, anxiety, ESL/ELL, etc.)
Phrase patterns for keyword extraction	100+ deterministic symptom/teacher-note patterns
Research sources scraped per plan	Up to 5 web pages per student query
Plan duration	2–3 weeks with 3–6 teacher action steps per segment
Seating algorithm dimensions	4 (academic, behavior, social, support needs)
Grades supported	K–12 (13 grade levels)
Student avatar options	16 emoji presets
Production dependencies	11 npm packages + 3 Python packages
Git branches	3 feature branches (classroom-grid, google-oauth, main)

---

What It Does

1. Student Profile Management

• Full CRUD for student profiles with issues, strengths, goals, behavioral notes, and avatar
• Grade-level filtering and search across K–12
• Per-teacher data isolation via Supabase RLS and JWT-scoped queries
• Real-time student list updates via Supabase Realtime subscriptions

2. AI Plan Generator

• 3-stage pipeline: keyword extraction → web research scraping → LLM plan synthesis
• Recognizes 12 special-education issue categories with 100+ deterministic phrase patterns (e.g., "easily distracted" → ADHD, "trouble decoding words" → dyslexia)
• Scrapes up to 5 research sources per student using Bing search + Playwright headless browser
• Generates structured 2–3 week intervention plans with weekly segments, teacher actions, student expectations, and progress check-ins
• Cites research sources with URLs in every generated plan
• Plans persist to Supabase with milestones, date ranges, and custom prompts

3. Classroom Seating Simulation

• Drag-and-drop seating grid with dynamic row/column resizing
• RBSB (Radius-Based Score Balancing) algorithm — a custom 305-line seating optimizer:
  • Composite scoring across 4 dimensions (academic 50%, behavior 30%, social 20%, support needs −40%)
  • Snake-pattern initial placement sorted by composite score
  • Iterative neighbor-swap balancing (up to 5 iterations) to minimize local/global imbalance
• One-click auto-sort, CSV export, analytics panel, and local persistence

4. Authentication & Security

• Google OAuth via Supabase Auth
• Protected routes on all 9 authenticated pages
• 12 Row-Level Security policies ensuring teachers only access their own classrooms, students, plans, and occurrences
• Teacher-scoped database view (teacher_students) for secure frontend queries

---

Architecture

consilai/                          # Monorepo root
├── apps/
│   ├── frontend/                  # Next.js 14 + React 18 + Tailwind CSS
│   │   ├── app/                   # 9 pages + 1 API route (App Router)
│   │   ├── components/            # 14 reusable UI components
│   │   └── lib/                   # Auth, Supabase client, seating algo, contexts
│   └── backend/
│       ├── src/services/          # RBSB seating algorithm (TypeScript)
│       └── supabase/migrations/   # 4 SQL migrations (530 LOC)
└── packages/
    ├── ai/                        # TypeScript AI pipeline (1,023 LOC)
    │   ├── keywordExtractor.ts    # 12 SPED categories, 100+ patterns (337 LOC)
    │   ├── researchFetcher.ts     # Spawns Python scraper subprocess
    │   ├── planGenerator.ts       # Azure Phi LLM integration
    │   └── prompts/               # Structured JSON plan prompts
    └── scraper/                   # Python research scraper (118 LOC)
        └── scrapers/scraper.py    # Bing + Playwright + BeautifulSoup

AI Pipeline Flow

Teacher selects student
        │
        ▼
┌─────────────────────┐
│ Keyword Extractor   │  12 issue categories, 100+ phrase patterns
│ (TypeScript)        │  Deterministic — no LLM needed
└─────────┬───────────┘
          │ search query (up to 8 terms)
          ▼
┌─────────────────────┐
│ Web Scraper         │  Bing search → 5 URLs → Playwright render
│ (Python)            │  BeautifulSoup text extraction
└─────────┬───────────┘
          │ ResearchSnippet[] with abstracts + summaries
          ▼
┌─────────────────────┐
│ Plan Generator      │  Azure Phi-3 LLM (temp 0.4)
│ (TypeScript)        │  Structured JSON: goals, segments, actions
└─────────┬───────────┘
          │
          ▼
   Plan + cited sources → Supabase

Database Schema

Table	Purpose	Key Fields
classrooms	Teacher-managed classrooms	name, description
students	Student profiles	name, grade, issues[], strengths[], goals[], seat position
plans	AI-generated learning plans	title, objectives, milestones[], date range
student_occurrences	Per-student AI interactions	prompt, ai_result (JSONB)
class_occurrences	Class-wide AI interactions	prompt, ai_result (JSONB)
student_occurrence_students	M2M join table	occurrence ↔ student links
teacher_students (view)	RLS-scoped student query	All student fields for current teacher

---

Tech Stack

Layer	Technology	Version
Frontend framework	Next.js (App Router)	14.x
UI library	React	18.x
Styling	Tailwind CSS (dark/light mode)	3.3
Icons	Lucide React	0.263
Auth	Supabase Auth + Google OAuth	—
Database	Supabase (PostgreSQL)	—
AI / LLM	Azure Phi-3	—
Research scraping	Python (Playwright, BeautifulSoup, Requests)	—
AI pipeline	TypeScript	5.4
Realtime	Supabase Realtime (Postgres changes)	—

---

Codebase Breakdown

Area	Files	Lines of Code	% of Total
Frontend (JSX pages)	10	~1,991	35%
Frontend (components)	14	~1,100	19%
Frontend (lib/utils)	8	~901	16%
AI package (TypeScript)	10	~1,023	18%
SQL migrations	4	~530	9%
Python scraper	1	~118	2%
CSS	1	~60	1%
Total	51	5,663	100%

Largest Modules

File	Lines	Description
app/classroom/page.jsx	670	Seating simulation with drag-and-drop, auto-sort, CSV export
app/plans/page.jsx	406	AI plan generator UI with Supabase persistence
migrations/0001_*.sql	401	Core schema, RLS policies, RPC functions
keywordExtractor.ts	337	12-category SPED keyword extraction engine
rbsbSeating.ts	305	Radius-Based Score Balancing seating algorithm
PlanResultCard.jsx	230	Plan display with milestones and source citations
StudentForm.jsx	193	Student profile form with validation
Navbar.jsx	179	Navigation with auth state and theme toggle

---

Key Features by the Numbers

• 10 application routes (home, login, auth callback, students list/create/detail/edit, classroom, plans, API)
• 14 reusable React components (Button, Desk, DeskGrid, StudentCard, PlanResultCard, etc.)
• 3 React context providers (Auth, Theme, Toast)
• 12 RLS policies protecting 6 database tables
• 2 Supabase RPC stored procedures for occurrence tracking
• 12 SPED/SEL issue categories with deterministic keyword expansion
• 100+ phrase patterns for symptom and teacher-note matching
• 5 web research sources scraped per plan generation
• 4 composite score dimensions in the seating algorithm
• 5 max balancing iterations in RBSB auto-sort
• 3–6 teacher action steps per plan segment
• 2–3 week plan duration windows
• 16 emoji avatar options for student profiles
• 13 K–12 grade levels supported in filtering
• 2 export formats (CSV for seating charts and plans)
• 2 theme modes (light and dark)

---

Getting Started

Prerequisites

• Node.js 18+
• Python 3.10+ with virtual environment
• Supabase project (URL + anon key)
• Azure Phi-3 deployment (endpoint, deployment name, API key)
• Google OAuth credentials (for Supabase Auth)

Environment Variables

Create apps/frontend/.env.local:

NEXT_PUBLIC_SUPABASE_URL=your_supabase_url
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key
AZURE_PHI_ENDPOINT=your_azure_phi_endpoint
AZURE_PHI_DEPLOYMENT=your_deployment_name
AZURE_PHI_API_KEY=your_api_key

Install & Run

# Frontend
cd apps/frontend
npm install
npm run dev          # http://localhost:3000

# AI package (TypeScript compilation)
cd packages/ai
npm install
npm run build

# Python scraper
cd packages/scraper
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Database migrations
# Apply SQL files in apps/backend/supabase/migrations/ via Supabase CLI or dashboard

Database Setup

Run migrations in order against your Supabase project:

1. 0001_classrooms_students_occurrences.sql — Core schema, RLS, RPC functions
2. 0002_students_profile_fields.sql — Student profile extensions (issues, strengths, goals)
3. 0003_plans.sql — AI plan persistence table
4. 0004_teacher_students_view_refresh.sql — View refresh for new columns

---

Project Structure

consilai/
├── apps/
│   ├── frontend/          # Next.js 14 teacher dashboard (~3,992 LOC)
│   └── backend/           # Supabase migrations + seating service (~835 LOC)
├── packages/
│   ├── ai/                # TypeScript AI pipeline (~1,023 LOC)
│   └── scraper/           # Python web research scraper (~118 LOC)
├── apps/database_schema.txt
└── README.md

---

Team & Development

Contributor	Commits
pho-muncher	30
manalaishabeer@gmail.com	13
Christian Chamberland	11
Tona	7

• 61 total commits over 27 days of active development
• 3 feature branches: classroom-grid, google-oauth, main
• Built collaboratively as a full-stack EdTech prototype

---

Resume Talking Points

"Built ConsilAI, a full-stack EdTech platform (~5,700 LOC) helping teachers generate AI-powered, research-backed learning plans for students with special needs."

• Architected a 4-package monorepo (Next.js frontend, Supabase backend, TypeScript AI pipeline, Python scraper) with 61 commits across 4 contributors in 27 days
• Designed a 3-stage AI pipeline: deterministic keyword extraction across 12 SPED categories and 100+ phrase patterns, automated web research scraping (5 sources/plan), and Azure Phi-3 LLM plan synthesis
• Implemented 12 Row-Level Security policies on 6 PostgreSQL tables with Google OAuth, ensuring complete per-teacher data isolation
• Built a custom RBSB seating algorithm (305 LOC) using composite scoring across 4 behavioral dimensions, snake-pattern placement, and iterative neighbor-swap balancing
• Delivered 14 React components across 10 routes with dark/light theming, drag-and-drop classroom simulation, real-time Supabase subscriptions, and CSV export
• Integrated Playwright headless browser scraping with BeautifulSoup text extraction, orchestrated from Node.js via subprocess spawning

---

License

Private project — all rights reserved.