Overview
Students who move to a new city face scattered information: where to live, which gym or mess is any good, and whom to talk to when things get hard. Service providers face the opposite problem. They rarely see honest, organised feedback from the students they serve.
Campus Heaven connects both sides. Providers list accommodations, gyms, libraries and meal services. Students browse, compare and leave star-rated reviews. A chatbot offers first-line support for common student concerns. Behind the scenes, every review and chat is analysed, and providers receive dashboards and a downloadable PDF report.
The problem
- Fragmented discovery: listings, prices and quality signals sit across word of mouth, social media and brokers.
- Little trustworthy feedback: reviews are informal, and sarcastic or mixed comments are hard to read at a glance.
- Unsupported students: homesickness, academic pressure and money worries often go unaddressed, especially for first-time movers.
- No insight for providers: owners cannot easily see ratings trends, recurring complaints or what students actually need.
The solution
Campus Heaven is a role-based web application with three connected modules and shared analytics. The design principle was simple: use lightweight, explainable components that can be built, deployed and maintained with modest resources.
| Module | What it does |
|---|---|
| Business | Providers register, then create, edit and delete listings across four categories (accommodation, gyms, libraries, meal services) with facilities, address, pricing and images. A personal dashboard shows performance. |
| Student | Browse listings by location, price and rating. Submit star ratings with written reviews. Every review is logged with listing, text, rating and timestamp. |
| “Urgent Care” chatbot | Retrieval-based assistant covering eight intent categories, such as homesickness, academic pressure and social isolation. Interactions are logged with the accommodation context. |
| Analytics and reports | Parses reviews and chat logs to compute average ratings, sentiment mix (positive, negative, neutral, sarcastic) and trending issues. Shows bar, pie and line charts and exports a PDF report. |
| Admin | Secure console to manage student and business accounts, moderate listings, and review or remove reviews and chat transcripts. |
How the AI works, in plain terms
The project deliberately avoids heavy models. Two techniques do most of the work:
- Chatbot intent matching: a student’s message and a set of example phrases for each intent are converted to word-count vectors (CountVectorizer). Cosine similarity picks the closest intent. If the best match scores below 0.3, the bot says it is unsure and asks for clarification instead of guessing.
- Review and sarcasm analysis: review text is cleaned, lowercased and tokenised, then classified into sentiment classes. A sarcasm check helps catch comments such as praise that is actually a complaint. The approach was informed by published work on context incongruity in sarcasm detection.
Because both methods are transparent, every chatbot reply and every sentiment label can be traced back to a clear rule or score. That matters for a tool that deals with student wellbeing.
Sample workflow
| Step | Example |
|---|---|
| Student input | Location: Nashik. Needs: accommodation and gym. Review: “Great room but a bit noisy at night.” Chat: “I’m feeling homesick, what should I do?” |
| System output | Suggested listings: PGs near college with gym access. Review sentiment: positive. Chatbot reply: an empathetic response encouraging peer connection and campus clubs. |
| Provider view | Report summary showing average rating and top concern (for example, noise), so the owner knows what to fix. |
Technology and architecture
The platform uses a three-tier design: presentation (Flask templates and JavaScript chat widget), application logic (Flask views organised as separate /student, /business and /admin blueprints), and data (SQLite plus text-based logs). Students, business owners and admins reach one Flask web app that handles role-based blueprints, authentication, listings, reviews and the chatbot API. It stores data in SQLite and text logs, which feed the analytics layer for sentiment, charts and PDF reports. Nginx and Docker sit in front for deployment.
| Technology | Why it was chosen |
|---|---|
| Python and Flask | Lightweight, easy to extend, and a rich ecosystem for security and reporting. |
| SQLite | Zero-configuration embedded database, ideal for a prototype and small-scale deployment. |
| scikit-learn | CountVectorizer and cosine similarity were enough for an explainable chatbot, with no need for large language models. |
| Pandas and NumPy | Parsing and aggregating review and chat data. |
| Matplotlib and Seaborn | Quick bar, pie and line charts, rendered headlessly for the server. |
| FPDF | Simple PDF report generation without heavy dependencies. |
| Git, Docker, Nginx | Version control with feature branches; containerised deployment with HTTPS and a reverse proxy. |
How it was built
- Requirement analysis: gathered needs from students, landlords and service providers, covering both features and qualities like security and ease of use.
- Design: ER diagrams, use case diagrams for each role, data flow diagrams (levels 0 to 2), a class diagram and wireframes for the dashboards.
- Development order: listing management first, then reviews and logging, then the chatbot API, then analytics layered on top of the logs.
- Testing: unit tests for the chatbot and sarcasm helpers, integration tests across register, list, review, chat and report flows, plus manual cross-browser checks and student trials.
- Deployment: Docker containers on a cloud VM, daily backups of the database and logs, and log rotation.
What was delivered
The project produced a complete, working prototype covering all three user roles end to end:
- Secure registration, login and role-based access for students, business owners and admins.
- Listing management across four service categories, with provider dashboards.
- Student browsing and star-rated reviews, stored for later analysis.
- A chatbot handling eight student-concern categories, with every interaction logged.
- Analytics with dynamic charts and a downloadable PDF report for providers.
- An admin console for moderation, account control and oversight.
A note on results. This case study reports what was built and how it works. Formal accuracy figures for the chatbot and sentiment classifier, and user-adoption numbers, were not part of the documented results, so none are claimed here.
Why this design works
| Design choice | Benefit |
|---|---|
| Explainable, rule-based AI | Predictable replies and traceable labels, which suits sensitive wellbeing topics and a small dataset. |
| Low-cost stack | Runs on a modest VPS with minimal setup, so it is realistic for a college or small business to adopt. |
| Everything is logged | Traceability for admins, and raw material for trend analysis of reviews and concerns. |
| Separate role modules | Each user sees only what they need, and the code stays easier to maintain. |
Limitations
- The chatbot is retrieval-based and covers a fixed set of intents. It does not hold open-ended conversations and is not a substitute for professional counselling.
- Sentiment and sarcasm detection use lightweight methods and may misread subtle or mixed language.
- SQLite and flat-file logs suit a prototype. Heavy concurrent use would need a stronger backend.
- Housing-demand forecasting, mentioned as a vision in the project introduction, remains future work.
Future scope
| Enhancement | Value |
|---|---|
| Real-time WebSocket chat | Instant two-way communication instead of request-and-poll. |
| Personalised recommendations | Collaborative or content-based filtering using student profiles and past reviews. |
| Housing demand forecasting | Predict demand by area and season using academic calendars and historical data. |
| Transformer-based NLP | Models such as BERT or RoBERTa for more nuanced intent and sentiment detection. |
| Mobile apps | Flutter or React Native apps with push notifications. |
| Multilingual and accessible UI | More languages and WCAG standards for international students and users with disabilities. |
| Scalable backend | Move to PostgreSQL or MongoDB with object storage for logs. |
| Campus system integration | Live room, library and meal-plan availability with single sign-on. |
Key learnings
- Choose the simplest technique that solves the problem. Cosine similarity was enough for a useful chatbot.
- Logging from day one made analytics, moderation and debugging much easier later.
- Designing for several user roles early kept the system organised as it grew.
- Honest reporting of limits builds more trust than inflated claims.
Conclusion
Campus Heaven shows how web development, data analysis and practical AI can be combined to solve a genuine everyday problem. It gives students clearer choices and support, gives providers usable insight, and gives administrators control. The same pattern of collecting data, analysing it simply and presenting clear actions carries over to many business settings.