AlumniConnect — Centralized Alumni Data Management & Engagement Platform SIH25017 Project Story & Technical Overview

💡 Inspiration Every year, thousands of students graduate from universities and enter the global workforce. However, after graduation, the connection between alumni and their alma mater often degrades into fragmented spreadsheets, static WhatsApp groups, or lost contacts. Current students miss out on invaluable career mentorship, industry placement referrals, and real-world guidance, while college management struggles to track alumni career progressions and aggregate institutional impact.

We were inspired to solve this structural disconnect by building AlumniConnect — a unified, centralized University SaaS platform. Our goal was to create a bridge connecting Students, Alumni, and College Administration, powered by an AI-driven mentor recommendation engine and executive Plotly analytics, ensuring every student finds the right mentor to accelerate their career.

🚀 What It Does AlumniConnect is a comprehensive digital ecosystem designed specifically for higher education institutions:

🎓 For Students:

AI Mentor Matching: Discover top alumni mentors recommended by an NLP similarity engine based on matching skills, academic department, and career aspirations. 1-on-1 Guidance Requests: Send personalized mentorship requests and track response statuses (Pending, Accepted, Completed). Exclusive Placements Board: Browse full-time job openings and AI-ranked internship opportunities posted directly by verified alumni. Campus Events & Webinars: Register for upcoming guest lectures, technical workshops, and annual alumni meets. 💼 For Alumni:

Professional Network Profile: Showcase career achievements, industry domain, current company, and toggle Mentorship Availability. Direct Referral Posting: Publish job referrals and student internship positions. Mentees Management: Review, accept, and guide student applications from a centralized dashboard. Achievements Showcase: Log publications, awards, and milestones into an institutional honor ledger. 🏛️ For College Administration:

Central Control Panel: Audit student and alumni databases with instant profile verification controls. Broadcasting Hub: Publish official college bulletins and send real-time system alerts to all users. Executive Plotly Analytics: 10 interactive data visualizations analyzing alumni distribution by department, graduation cohorts, industry sectors, location breakdown, and platform engagement growth trends. 🧮 How We Built It & The Math Engine Behind AlumniConnect AlumniConnect was built using a lightweight, performant tech stack:

Backend Framework: Python 3, Flask, SQLite3, SQLAlchemy ORM, Flask-Login. Frontend UI/UX: HTML5, CSS3, JavaScript, Tailwind CSS CDN, FontAwesome 6. Data & Analytics Engine: Pandas, NumPy, Plotly.js. Machine Learning & NLP Pipeline: NLTK, Scikit-Learn (TF-IDF Vectorizer & Cosine Similarity). The Math: AI Mentor & Internship Recommendation Engine To match a student s s with the most relevant alumni mentor a a, we represent profiles as high-dimensional feature vectors in a term-vector space built from technical skills, career goals, interests, and academic majors.

1. Term Frequency-Inverse Document Frequency (TF-IDF) Weighting For a skill or term t t in a profile document d d within the profile corpus D D, Term Frequency T F ( t , d ) TF(t,d) measures how frequently the term appears in document d d:

T F ( t , d

)

f t , d ∑ t ′ ∈ d f t ′ , d TF(t,d)= ∑ t ′ ∈d ​ f t ′ ,d ​

f t,d ​

​

Inverse Document Frequency I D F ( t , D ) IDF(t,D) penalizes common terms that appear across all profiles:

I D F ( t , D

)

ln ⁡ ( 1 + ∣ D ∣ 1 + ∣ { d ∈ D : t ∈ d } ∣ ) + 1 IDF(t,D)=ln( 1+∣{d∈D:t∈d}∣ 1+∣D∣ ​ )+1 The composite TF-IDF weight for term t t in profile d d is given by:

w t ,

d

T F ( t , d ) × I D F ( t , D ) w t,d ​ =TF(t,d)×IDF(t,D)

1. Vector Cosine Similarity Score We construct normalized feature vectors v s v s ​ for the student and v a v a ​ for each candidate alumni mentor. The raw mathematical similarity score S raw ( s , a ) S raw ​ (s,a) is computed using the dot product normalized by Euclidean norm ( L 2 L 2 ​ norm):

cos ⁡ ( θ

)

v s ⋅ v a ∥ v s ∥ 2 ∥ v a ∥

2

∑

i

1 N w i , s w i , a ∑

i

1 N w i , s 2 ∑

i

1 N w i , a 2 cos(θ)= ∥v s ​ ∥ 2 ​ ∥v a ​ ∥ 2 ​

v s ​ ⋅v a ​

​ = ∑ i=1 N ​ w i,s 2 ​

​

∑ i=1 N ​ w i,a 2 ​

​

∑ i=1 N ​ w i,s ​ w i,a ​

​

1. Domain & Department Bonus Boost To ensure academic context is preserved, we apply a weighted domain boost B ( s , a ) B(s,a) based on department alignment:

⚠️ Failed to render LaTeX: KaTeX parse error: Expected 'EOF', got '&' at position 6: 0.15 &̲ \text{if } \te… 0.15 & \text{if } \text{Dept}(s) = \text{Dept}(a) \ 0.00 & \text{otherwise} \end{cases}$$

The final match score percentage $M(s, a)$ is calculated as:

$$M(s, a) = \min \left( 100, \, \left( 0.85 \times \cos(\theta) + B(s, a) \right) \times 100 \right)$$

---

⚡ Challenges We Faced

1. Handling Profile Data Sparsity:

   • Challenge: Newly registered students often have brief profile descriptions, making vector representations sparse and leading to low raw cosine similarity scores.
   • Solution: We implemented dynamic fallback heuristics and explainable rationale generation (e.g., matching on preferred industry and department overlap when skill overlap is low).

2. Creating a Luxury SaaS UI Without Heavy JS Frameworks:

   • Challenge: The brief mandated using standard HTML5/CSS3/JavaScript without heavy single-page application frameworks like React or Vue, while maintaining a modern, responsive light-theme SaaS visual experience.
   • Solution: We built a custom utility CSS system on top of Tailwind CSS CDN, implementing glassmorphism headers, card hover-lift effects, animated tabs, and interactive drawer controllers in vanilla JavaScript.

3. Ensuring High Template Compatibility with Flask Routes:

   • Challenge: Redesigning over 40 Jinja2 templates without breaking existing backend route variables, form action endpoints, or SQLite database models.
   • Solution: We conducted rigorous automated testing using a custom test suite (python test_app.py) to verify HTTP response statuses, form submissions, and Plotly chart renders after every UI edit.

---

🎓 What We Learned

• NLP & Vector Space Modeling: Gained hands-on experience structuring text preprocessing pipelines using NLTK tokenization, TF-IDF vectorization, and cosine similarity for real-time recommendation systems.
• Modern SaaS UI System Architecture: Mastered light-theme color theory (#F7F9FC background, #2563EB royal blue, #0F9D8A alumni teal, #7C3AED admin violet) and responsive layout patterns.
• Interactive Data Visualization with Plotly: Learned how to generate 10 dynamic, responsive Plotly charts server-side with Pandas and embed them seamlessly inside Flask templates.

---

🔮 What's Next for AlumniConnect

• Real-time WebSockets Chat: Implementing direct 1-on-1 instant messaging between accepted mentees and alumni mentors.
• Automated LinkedIn Profile Import: Integrating LinkedIn OAuth API to allow alumni to auto-populate their experience, skills, and current company in one click.
• Mobile App Edition: Developing native iOS and Android apps using React Native powered by Flask RESTful APIs.

Built With

• css3
• data-analytics
• data-visualizationnltk
• edtech
• flask
• fontawesome
• html5
• javascript
• machine-learning
• natural-language-processing
• numpy
• pandas
• python
• responsive-designplotly
• scikit-learn
• sqlalchemy
• sqlite
• tailwindcss
• tfidf
• vector-cosine-similaritysaas