Dakhla – AI-Assisted University Admissions Platform

Overview

Dakhla is an AI-assisted university admissions platform developed as my Final Year Project.

The platform was designed to simplify the university discovery and application process by bringing program search, recommendations, AI assistance, document handling, and application management into a single system.

It supports multiple user roles, including students, administrators, and universities, with separate workflows for partner and non-partner institutions.

Problem

University admissions information is often fragmented across many websites, application portals, PDFs, and institutional pages.

Students may need to manually compare universities, programs, eligibility requirements, deadlines, and admission criteria across multiple sources.

At the same time, universities need structured ways to receive, review, filter, and manage applications.

Dakhla was designed to centralize these workflows while adding AI-assisted guidance and recommendation features.

Platform Architecture

The platform uses a full-stack architecture built with:

• React
• Tailwind CSS
• FastAPI
• Supabase
• PostgreSQL

The frontend provides university discovery, program exploration, application workflows, and AI-assisted interactions.

The backend handles authentication, role-based access control, recommendation services, document processing, and application-related APIs.

User Roles

Dakhla supports multiple user types:

Students

Students can:

• Search and explore universities
• Filter programs
• View university and program details
• Receive AI-assisted recommendations
• Interact with the admissions chatbot
• Submit applications
• Upload supporting documents

Universities

Partner universities receive additional functionality for managing applications and collecting institution-specific information.

Universities can review submitted applications and work with application data through structured workflows.

Administrators

Administrators manage platform-level data and system functionality.

AI Admissions Chatbot

Dakhla includes an AI-powered chatbot designed to assist students with admissions-related questions.

The chatbot uses Groq with Llama 3.3 70B and a Retrieval-Augmented Generation workflow.

The RAG pipeline can work with university-related information and uploaded content, allowing the assistant to answer questions using available admissions context instead of relying only on general model knowledge.

The ingestion workflow supports multiple content types, including:

• PDFs
• Documents
• Images
• Web-based university content

This allows the system to build a broader admissions knowledge base for student queries.

University Recommendation System

Dakhla also includes a machine learning recommendation component.

A Random Forest classifier was trained on a dataset containing 5,000 student records and 14 program categories.

The initial model achieved approximately 78.5% accuracy.

After hyperparameter tuning using GridSearchCV across 108 configurations with 5-fold cross-validation, the model achieved approximately 82.7% accuracy.

The recommendation system is intended to support program discovery by matching student-related inputs with suitable academic program categories.

University Discovery

The platform provides a centralized university directory with search and filtering capabilities.

University information is combined with structured platform data to provide students with a single interface for exploring institutions and academic programs.

Partner University Workflows

Partner universities can define custom questions that students must answer during the application process.

This allows the application form to adapt to institution-specific requirements rather than forcing every university into one fixed form structure.

Application Management

University users can review submitted applications through an administrative interface.

The application management workflow supports:

• Viewing applications
• Sorting
• Filtering
• Accepting or rejecting applications
• Selecting fields for export
• Exporting application data

Export options were designed to support formats such as CSV, Excel, and structured API-based access.

Authentication and Access Control

The platform uses JWT-based authentication and role-based access control.

Access to functionality is determined by the authenticated user's role, allowing students, university users, and administrators to interact with different parts of the system.

Automated Data Collection

Dakhla also includes automated university data collection workflows.

Firecrawl and APScheduler were used to support recurring retrieval and updating of university-related information.

This helped reduce reliance on manually maintaining every piece of university content.

What I Learned

Dakhla was one of my most comprehensive projects because it combined AI, machine learning, backend development, database design, authentication, document processing, data ingestion, and frontend development in one platform.

The project gave me hands-on experience designing a multi-role product where AI features had to operate alongside real application workflows rather than as isolated demos.