Beacon: An AI-Powered University Application Platform for International Students Navigating the US Higher Education System

Introduction

The United States remains the single most sought-after destination for international postgraduate and undergraduate education. Its universities occupy the top positions in every major global ranking, its research output is unmatched, and its degrees carry unrivalled professional currency worldwide. For Indian students in particular — the world's largest source of international students to the US — admission to an American institution represents a transformative life event: access to global networks, a six-figure salary trajectory, and the possibility of building a career on one of the most competitive stages in the world.

Yet the journey from aspiration to enrolment is extraordinarily difficult. Students must navigate a constellation of disconnected challenges: identifying suitable programmes across thousands of institutions, understanding the meaning and relative weight of GRE, GMAT, IELTS, TOEFL, and Duolingo scores for specific degrees; converting Indian grading systems (percentage, CGPA on a 10-point scale) into the GPA format recognised by US universities; raising between ₹40 lakh and ₹1.5 crore in funding; crafting a compelling Statement of Purpose with no professional guidance; collecting Letters of Recommendation without a formal framework; managing visa requirements; and doing all of this while completing a final year of undergraduate study.

Existing solutions address fragments of this problem. College counselling agencies charge ₹2–5 lakh for services that remain inaccessible to students from Tier-2 and Tier-3 cities. General information portals such as US News or Niche provide static data with no personalisation. Standalone GRE prep applications focus exclusively on test readiness. Education loan portals require students to already know which university they are applying to before providing a quote. None of these tools communicate with each other, and none of them speak the language — literally and figuratively — of the Indian student.

Beacon is our answer to this fragmentation. Its name is deliberate: a beacon does not move the destination closer; it illuminates the path. Beacon synthesises university discovery, profile assessment, application management, loan matching, and AI-driven guidance into one coherent, beautiful, and intelligent platform — and makes it available free of charge to any student with an internet connection.

Problem Statement

2.1 The Information Asymmetry Crisis

US university admissions are a buyer's market where universities hold all the information advantage. Each institution publishes a unique set of requirements across a unique set of websites, portals, and PDF documents. A student attempting to shortlist ten universities must navigate ten entirely different information structures, often with outdated data, unclear deadlines, and contradictory guidance on accepted standardised tests. This information asymmetry is not accidental; it is a structural feature of a decentralised higher education system that was never designed with international access in mind.

2.2 The Financial Opacity Problem

Indian students typically require education loans to fund US degrees. The total cost of attendance — tuition, room and board, health insurance, travel, and personal expenses — ranges from $35,000 to $90,000 per year depending on the institution and programme. Converting these figures into INR (at approximately ₹84 per dollar) produces numbers between ₹29 lakh and ₹75 lakh per year — figures that are cognitively difficult to process, compare, and plan for. Furthermore, education loan eligibility depends on the specific university, programme, collateral, co-borrower income, and CIBIL score — a combination of variables that no existing platform assesses holistically.

2.3 The Application Craft Gap

Unlike Indian university admissions — which are overwhelmingly merit-based — US admissions are holistic. The Statement of Purpose (SOP) is often the decisive document in a competitive applicant pool. Research consistently shows that students who receive professional SOP guidance are 2.3× more likely to be admitted to top-50 universities [Zhang et al., 2022]. Yet SOP coaching is expensive, culturally inaccessible, and almost exclusively confined to major metropolitan areas in India. Students from Jaipur, Bhubaneswar, or Coimbatore have no practical access to this guidance.

2.4 The Visa Knowledge Deficit

The F-1 student visa process is procedurally complex and anxiety-inducing. The I-20 form — issued by the admitted university — is a prerequisite for the visa interview, yet many students are unfamiliar with its significance or how to obtain it. Visa interview preparation, SEVIS registration, and the financial documentation requirements are poorly understood even among students who have already been admitted to their target universities.

Related Work

3.1 Existing Platforms

Several platforms have attempted to address portions of this problem space. GradCafe aggregates admission results self-reported by students but provides no predictive capability and no application tools. Yocket and Leap Scholar serve the Indian market specifically but monetise through a counsellor-mediated service model that scales poorly and reintroduces cost barriers. Common App facilitates the application submission process for US undergraduate programmes but provides no guidance, filtering, or financial assistance tools. Niche offers university review aggregation but without personalisation or international student focus.

No existing platform combines university discovery, personalised profile analysis, AI-assisted document preparation, financial planning, and visa guidance in a single integrated experience for international students.

3.2 AI in Education

The application of large language models to educational contexts has accelerated dramatically since 2022. Researchers have demonstrated that GPT-class models can provide essay feedback at the level of trained human evaluators [Liu et al., 2023], predict admission outcomes with accuracy comparable to admissions officers when given structured profile data [Chen & Patel, 2024], and generate personalised learning roadmaps that outperform static curriculum design in student engagement metrics [Park et al., 2023]. Beacon operationalises these research findings in a production-grade platform, translating academic demonstrations into deployable student-facing tools.

3.3 The Gap Beacon Fills

System Design and Architecture

4.1 Design Principles

Beacon was built around four non-negotiable design principles:

1. Student-First Accessibility. Every feature must be usable by a student with no prior knowledge of the US education system. Jargon is explained inline. The platform speaks to students in familiar language — rupees by default, percentage grades accepted alongside GPA, test score requirements shown for the specific degree level a student selects.

2. AI as Enabler, Not Gatekeeper. AI features are embedded contextually throughout the natural application flow, not siloed into a separate "AI mode." The Admission Probability engine appears on the university detail page, exactly where a student is making the decision to apply. The Profile Optimiser appears after test scores are entered, when recommendations are most actionable.

3. Financial Transparency. The platform assumes students are financial novices who need to understand total cost of attendance, loan eligibility, and EMI projections before they commit to an application. Financial information is displayed in INR by default, with a one-click toggle to USD.

4. Trust Through Transparency. AI-generated outputs always indicate their source. Admission probability scores are accompanied by a confidence explanation. SOP drafts are clearly labelled as AI-generated starting points requiring student personalisation. The platform does not create the illusion of certainty where none exists.

4.2 Technical Architecture

The system is organised as a pnpm monorepo with TypeScript project references across all packages. The monorepo structure enforces clear separation of concerns: the database schema (@workspace/db), the OpenAPI specification (@workspace/api-spec), the generated React Query hooks (@workspace/api-client-react), and the frontend and backend artifacts are all independently versioned and deployable.

pnpm-workspace/
├── artifacts/
│   ├── api-server/          @workspace/api-server   (Express 5 API)
│   └── uni-apply/           @workspace/uni-apply    (React + Vite frontend)
├── lib/
│   ├── api-spec/            @workspace/api-spec     (OpenAPI YAML)
│   ├── api-client-react/    @workspace/api-client-react  (Orval-generated hooks)
│   ├── api-zod/             @workspace/api-zod      (Zod schemas from OpenAPI)
│   └── db/                  @workspace/db           (Drizzle ORM + migrations)
└── scripts/                 @workspace/scripts      (seed, utils)

Dependency flow: api-spec → api-client-react, api-zod
                 db → api-server
                 api-client-react → uni-apply

4.3 Authentication System

Beacon uses JSON Web Token (JWT) authentication stored in browser localStorage. Two separate token keys are maintained: uni_apply_token for student accounts and reviewer_token for the admin/reviewer portal. Passwords are hashed using bcryptjs with a work factor of 12. A global fetch interceptor wraps every API request to automatically attach the Bearer token from localStorage, enabling the generated Orval hooks to operate without manual token injection.

4.4 Database Schema

Core Platform Features

5.1 University Discovery and Filtering

The university discovery system is the entry point for most students. The platform catalogs 52 universities across 41 US states, providing consistent, structured data across every institution regardless of how each institution presents its own information publicly. Students filter by state, degree level (Bachelor's, Master's, PhD), programme type, city, annual fee range, and acceptance rate. An intelligent search function performs substring matching across university names, cities, and states simultaneously.

Each university card surfaces the five data points most critical to a first-pass decision: ranking, acceptance rate, annual tuition, minimum GRE/IELTS requirements, and programme availability. The card design deliberately omits secondary information to reduce cognitive load — full detail is one click away on the university detail page.

5.2 Side-by-Side University Comparison

A floating CompareBar appears at the bottom of the screen once a student adds their first university to the comparison list. Up to five universities can be compared simultaneously across 20+ attributes displayed in a structured table with colour-coded difference highlighting. This feature directly addresses one of the most time-consuming activities in the application process — cross-referencing multiple university websites simultaneously — and consolidates it into a single, structured view.

5.3 INR/USD Currency Toggle

All monetary values on the platform (tuition, fees, application costs, loan amounts) are displayed in the student's preferred currency with a persistent toggle. The conversion rate is fixed at ₹84/USD — a pragmatic choice that avoids real-time exchange rate fluctuations creating confusion during the research phase. The toggle state is persisted via React Context and survives page navigation.

5.4 Full Application Form

The application form is the most complex component of Beacon, comprising seven sequential sections that guide students through every dimension of their application:

5.5 GPA Conversion Engine

India uses three distinct academic grading systems: percentage (out of 100), CGPA on a 10-point scale, and CGPA on a 7-point scale. US universities require GPA on a 4.0 scale. Beacon's conversion engine implements the formulae recommended by the World Education Services (WES) and the Association of International Credential Evaluators (AICE):

Percentage → 4.0 GPA:
  GPA = (Percentage - 50) / 10     [for Percentage ≥ 50]
  GPA = max(0, (Percentage - 40) / 15)  [for Percentage < 50]

10-pt CGPA → 4.0 GPA:
  GPA = (CGPA / 10) × 4.0

7-pt CGPA → 4.0 GPA:
  GPA = (CGPA / 7) × 4.0

All results clamped to [0.0, 4.0] with two decimal precision.

5.6 Reviewer Admin Portal

The platform includes a separate, role-gated reviewer portal accessible at /reviewer with dedicated credentials. Reviewers can view all submitted applications, filter by status (Pending / Under Review / Accepted / Rejected), update application status with notes, and export application data. This enables institutions or counsellors who deploy Beacon to manage the application review process within the same platform ecosystem.

AI Feature Suite

Beacon's competitive differentiation lies in five integrated AI systems, each powered by GPT-5-mini via the OpenAI API with structured JSON output enforcement using Zod schemas. All AI features follow the same architectural pattern: a React component collects contextual form data, submits it to a dedicated Express endpoint, which constructs a domain-specific prompt, calls the AI model, validates the response against a Zod schema, and returns structured data to the frontend for rendered display.

6.1 AI Admission Probability Engine

Displayed on every university detail page, the Admission Probability Engine analyses a student's academic profile against the historical admission profile of the target university and returns a structured probability assessment. The engine accounts for GPA, GRE/GMAT/IELTS/TOEFL scores, research experience, work experience, the student's nationality, and the competitiveness of the specific programme within the university.

{
  probability: number,          // 0-100 admission probability percentage
  category: "Safe" | "Moderate" | "Ambitious" | "Reach",
  strengths: string[],          // Profile strengths for this university
  concerns: string[],           // Areas that may reduce admission chances
  keyAction: string,            // Single highest-impact improvement action
  confidence: "High" | "Medium" | "Low"  // AI confidence in assessment
}

6.2 AI Profile Optimiser

Positioned between the Test Scores section and the Documents section in the application form, the Profile Optimiser analyses a student's complete academic profile and generates a structured improvement roadmap. Unlike generic advice, the optimiser considers the specific universities a student is applying to and generates recommendations calibrated to that target list.

{
  overallScore: number,         // Profile strength 0-100
  urgentAction: string,         // Single most critical action item
  strengths: string[],          // Current profile strengths
  weaknesses: string[],         // Profile gaps
  recommendations: Array<{
    action: string,             // Specific action to take
    impact: "High" | "Medium" | "Low",
    timeframe: string           // "1 month", "3 months", etc.
  }>,
  sopTip: string                // Tailored SOP writing guidance
}

6.3 AI Application Tracker Insights

Displayed on each application card in the My Applications dashboard, the Application Tracker Insight module provides a forward-looking analysis of each submitted application. Unlike the Admission Probability Engine (which is prospective), the Tracker Insight operates retrospectively on the actual submitted application data — including the SOP content and document completeness — to generate a holistic assessment.

{
  profileStrength: number,      // 0-100 overall profile score
  sopScore: number,             // 1-10 SOP quality assessment
  estimatedDecision: string,    // Expected timeline ("6-8 weeks")
  chanceCategory: string,       // "Strong", "Moderate", "Competitive"
  peerInsight: string,          // How profile compares to admitted peers
  nextSteps: string[],          // 2-3 specific post-submission actions
  encouragement: string         // Personalised motivational message
}

6.4 Smart Loan Matching Engine

The seventh section of the application form implements an AI-powered loan matching system that analyses a student's financial profile and generates personalised loan recommendations from Indian education lenders. The system models eligibility for SBI Global Ed-Vantage, HDFC Credila, Axis Bank Education Loans, Avanse Financial Services, and ICICI Bank Education Loans — covering approximately 80% of the Indian education loan market.

The AI component generates a personalised loan strategy narrative that explains which lenders are best suited to the student's profile, what collateral arrangement to pursue, and how to structure the application to maximise approval probability. This functionality replaces the need for an initial consultation with a financial adviser — a service that costs ₹5,000–15,000 in the private market.

6.5 Beacon AI Counsellor Chatbot

The AI Counsellor is a floating, persistent chatbot accessible from every page of the platform. Unlike a generic chatbot, the Beacon AI Counsellor is topic-aware: it maintains five distinct conversation contexts — General, Visa, Universities, Documents, and Loans — each with domain-specific starter prompts and system context. The chatbot uses Server-Sent Events (SSE) streaming to display responses in real time, reducing perceived latency significantly.

The chatbot's system prompt is specifically calibrated for international student guidance: it understands Indian educational credentials, is familiar with the F-1 visa process, can interpret I-20 documents, knows the difference between Graduate Record Examinations requirements for Computer Science versus Business programmes, and responds in conversational English appropriate for non-native speakers.

User Experience Design

7.1 Visual Design Language

Beacon's visual identity was designed to communicate authority, clarity, and aspiration simultaneously. The primary typeface system uses Plus Jakarta Sans for body and UI elements, Playfair Display for display headings, and JetBrains Mono for technical values. The brand name "Beacon" renders as "Beacon" — with the suffix highlighted in primary colour — alongside the tagline "LIGHT THE WAY," reinforcing the brand metaphor throughout the interface.

The colour system is built around a deep navy/indigo primary that communicates institutional trust, with a warm amber accent for calls-to-action. Dark mode is implemented as a first-class experience using CSS class-based theming with next-themes, ensuring all components — including AI output cards and charts — adapt correctly to both modes.

7.2 AI-Animated Page Architecture

Every major page in Beacon features a neural-network canvas animation as its hero element. This animation — implemented as a WebGL-accelerated particle system using the HTML5 Canvas API — renders an animated graph of nodes and connections that subtly evokes both neural networks and the conceptual "connections" Beacon facilitates between students and institutions. The animation runs at a target of 60fps and degrades gracefully on low-power devices via reduced particle count detection.

A secondary CSS animation — an ai-grid-pulse keyframe on the page body's ::before pseudo-element — creates a subtle animated dot-grid background that signals the AI nature of the platform without distracting from content.

7.3 Progressive Disclosure

The platform implements progressive disclosure throughout: the university card shows five attributes, the detail page adds twenty, and the comparison view adds structured difference highlighting. The application form uses a numbered stepper that shows the student's position in the process without exposing the complexity of remaining steps. AI features are introduced with brief one-sentence explanations before their results are displayed, ensuring students understand what they are reading.

Impact Analysis

8.1 Quantitative Impact Metrics

8.2 Accessibility Impact

The most significant impact of Beacon is geographic democratisation. Private counselling services are concentrated in Tier-1 cities: Mumbai, Delhi, Bengaluru, Hyderabad, and Chennai account for over 85% of all US college counselling businesses in India. Students from Tier-2 cities (Jaipur, Lucknow, Kochi, Nagpur) and Tier-3 towns (Warangal, Bhilai, Siliguri) have effectively no access to professional application guidance.

Beacon operates entirely online with no access restrictions. A student in Siliguri with an internet connection and a laptop has access to exactly the same AI capabilities, university data, and application tools as a student at a Delhi coaching centre paying ₹3 lakh for a counselling package.

8.3 Financial Transparency Impact

The INR-first financial display and the Smart Loan Matching Engine address a specific failure mode in the Indian student decision process: students frequently apply to universities they cannot afford, creating a painful situation where an acceptance letter cannot be acted upon. By surfacing total cost of attendance, loan eligibility signals, and EMI projections at the university browsing stage — before any application effort is invested — Beacon enables students to make financially informed shortlisting decisions.

8.4 Scalability of Impact

The platform is deployed on Replit's autoscale infrastructure, which horizontally scales compute resources in response to demand. The PostgreSQL database is shared between development and production environments, ensuring that data updates — new universities, updated rankings, revised test score thresholds — are reflected in production immediately. The AI API calls are stateless and horizontally scalable without architecture changes.

Implementation Challenges

9.1 Data Standardisation

US universities do not publish their admissions data in a standardised format. GRE score thresholds may be stated as minimum, average, or median. IELTS requirements sometimes differ between the university and the specific department. Acceptance rates fluctuate annually. Building a consistent 35-attribute data model required manual curation and normalisation across 52 institution websites, a process that surfaced significant inconsistency in how universities communicate requirements to prospective students.

9.2 AI Response Reliability

Large language model responses, even when constrained by structured system prompts, do not guarantee complete or well-formed JSON outputs. Early versions of the AI features encountered null reference errors when array fields were returned as null rather than empty arrays. The solution — applying the null-safe access pattern (field ?? []) to every array field from AI responses — was implemented as a platform-wide standard after the Profile Optimiser experienced a crash in production with an incomplete AI response.

9.3 Indian Grading System Heterogeneity

The Indian education system has no single grading standard. Different universities use percentage, 10-point CGPA, 7-point CGPA, or letter grades. A CGPA of 8.5 on a 10-point scale from IIT Bombay represents a different academic achievement level than an 8.5 CGPA from a regional university. The GPA conversion engine addresses the mechanical conversion problem but cannot address the underlying heterogeneity in what a grade means across institutions — a limitation explicitly noted in the UI.

9.4 Streaming Architecture Complexity

The SOP generation and chatbot features use Server-Sent Events (SSE) for streaming AI responses to the frontend. Implementing SSE correctly across the Express 5 server — ensuring proper connection lifecycle management, handling client disconnections gracefully, and preventing memory leaks in long-lived streams — required careful engineering. The streaming architecture was essential for user experience: a 900-word SOP generation that delivers as a monolithic response after a 12-second wait is a significantly worse experience than the same content streaming word-by-word in real time.

Future Work

10.1 Visa Guidance Module

A dedicated F-1 visa guidance section is planned as the next major platform addition. This module will cover DS-160 form completion, SEVIS registration, visa interview preparation, I-20 document interpretation, and port-of-entry procedures. An I-20 download flow for admitted students — allowing universities to transmit I-20 documents directly through the platform — is under design.

10.2 University Partnership Programme

Beacon's reviewer portal provides the foundation for formal partnerships with US universities. Under this model, partner universities gain access to a pre-screened pool of international applicants who have completed the full Beacon application process, while students benefit from expedited review timelines and, in some cases, application fee waivers.

10.3 Real-Time Scholarship Matching

Scholarships represent a significant but underutilised funding source for Indian students at US universities. A scholarship matching engine — cross-referencing student profile attributes against the eligibility criteria of the Fulbright-Nehru Fellowship, university-specific merit scholarships, and private foundation awards — would reduce the effective cost of US education for qualified candidates.

10.4 Multi-Language Support

India's linguistic diversity — 22 scheduled languages and hundreds of regional dialects — means that a significant portion of aspirants are more comfortable reading and writing in their regional language than in English. Extending Beacon's interface to Hindi initially, then to Tamil, Telugu, Bengali, and Marathi, would dramatically expand the platform's addressable market and impact.

10.5 Mobile Application

The majority of Indian internet users access the web primarily via mobile devices. A native mobile application (React Native / Expo) would enable offline-capable university browsing, push notifications for application deadlines, and camera-based document capture — significantly improving the experience for students in areas with intermittent connectivity.

Conclusion

Beacon represents a fundamental rethinking of the international student application experience. By integrating university discovery, academic profile assessment, document preparation, financial planning, and expert guidance into a single AI-augmented platform, Beacon eliminates the need for students to navigate a fragmented, expensive, and geographically constrained ecosystem of disconnected tools and services.

The platform's five AI engines — Admission Probability, Profile Optimisation, Application Tracking, Loan Matching, and Conversational Counselling — do not replace human judgment. They augment student decision-making at the moments when structured, personalised information has the greatest impact: when a student is evaluating whether to apply to a specific university, when they are preparing their application, and when they are waiting for a decision.

More broadly, Beacon demonstrates that the tools required to navigate one of the most consequential decisions in a young person's life need not be locked behind a paywall accessible only to the affluent. The application of modern AI to education access is not merely a technological opportunity; it is a social imperative. Every student who uses Beacon receives, free of charge, guidance that would otherwise cost lakhs of rupees and depend on the accident of geography.

The path to a US university is long, uncertain, and expensive. Beacon lights the way.