ConnectEd

Powered by AI.Driven by humans. Starting at GMU, scaling mentorship to every campus.

Comment

Inspiration

Last semester, I reached out to 47 Google engineers on LinkedIn hoping for career advice. I heard back from just 3 people and it took 11 days. That’s a 6.4% response rate, and it showed me how hard meaningful networking really is for students.Everything changed at HackFax x PatriotHacks. Within 30 minutes, I met 3 Microsoft engineers from GMU, got resume feedback, and found a study partner. In-person connection removed all friction. We realized: what if AI could recreate that hackathon magic every day? But here's the key,AI shouldn't replace connections. ChatGPT can't review your resume with empathy or share what Google interviews are really like. AI should connect people to people, not replace them. That became our philosophy.

What it does

ConnectEd is a smart networking platform built exclusively for GMU students that tackles two major challenges: finding the right career mentor and getting academic help when you need it. In the Career Stream, imagine typing "I want to talk to someone who works at Google in software engineering" and instantly seeing alumni who match exactly what you're looking for. Each person comes with a match score and a clear explanation like "You're both Computer Science majors, you both love hiking, and they work at your dream company." The app even helps you write a friendly introduction to break the ice, so no more awkward cold messages. In the Student Stream, need a study buddy for your hardest class? Just type "I need help with CS 471" and find tutors who aced that exact course. Looking for a gym partner? Search "I need a swimming partner" and meet students who share your interests. When you post a help request, the platform automatically finds the best-fit tutors based on their grades and availability. Once you connect with anyone, you can chat directly in the app. Every suggestion comes with context so you know why you're a good match, making it easy to start genuine conversations. It's like having a personal assistant who knows everyone at GMU and can instantly introduce you to exactly the right person.

How we built it

We built ConnectEd with a Python FastAPI backend and React frontend, using three AI providers (Groq, Ollama) for intelligent matching. Our key innovation was a hybrid database design: SQL columns for fast filtering (major, company) and JSON blobs for AI-ready profile data. This let us query efficiently and match intelligently in one pipeline. Our three-stage matching pipeline extracts intent from natural language ("I need a swimming partner"), filters candidates via hybrid keyword/field matching, then computes scores combining regex frequency (50%) and LLM semantic ranking (50%). For career matching, we added weighted signals: major overlap (40%), target company (30%), shared hobbies (15%), and career goal alignment (15%). The "2-second magic" tutor dispatch is our standout feature. When students post help requests, we parse the course, filter tutors via indexed SQL, rank them with parallel LLM calls, and return top 3 matches with explanations, all in under 2 seconds. We optimize with parallel processing, cached explanations, and careful indexing. During demos, we show this live with two screens: student posting, tutor receiving instant notification.

Challenges we ran into

Building ConnectEd presented several technical hurdles. Implementing LinkedIn-style chat popups with persistent state across page navigation while ensuring secure logout data cleanup was complex. Balancing our AI-powered matching logic required creating an efficient system combining rule-based scoring with Groq's LLM for semantic matching, ensuring both speed and accuracy. Database integrity issues arose when handling connections between deleted users, requiring careful null checks and cascade deletes. Resume parsing was challenging: extracting meaningful text from multiple formats (PDF, DOCX, TXT) and storing them as base64 blobs in SQLite while handling encoding edge cases. Finally, crafting seamless authentication that prevented unauthorized access without disrupting user experience required strategic auth checks throughout our component architecture.

Accomplishments that we're proud of

We built a complete end-to-end networking platform with authentication, profiles, connections, messaging, and AI matching within a hackathon timeline. Our intelligent resume integration extracts text from multiple formats and integrates it into AI matching algorithms, awarding bonus points when resumes mention target companies. The messaging system features professional chat interfaces with minimize/maximize functionality and real-time polling with permanent SQLite storage. Our dynamic AI matching engine, powered by Groq's Llama 3.3 70B, delivers intelligent tutor matching with reasoning, semantic alumni search, and personalized connection message drafting. We're especially proud of our clean, scalable architecture with modular backend services and React frontend using Zustand state management, making the codebase maintainable and extensible beyond the hackathon

What we learned

We learned so much building this in such a short time! Working with AI was humbling, we spent hours figuring out how to get LLMs to actually return clean JSON instead of random text, and learned the hard way that you need backup plans when APIs fail (which they do, a lot). The coolest discovery was that mixing simple keyword matching with AI magic actually works better than pure AI alone. Running Ollama locally was a game-changer , we could do unlimited AI searches without worrying about API costs or rate limits. No more "we hit our API limit" panic at 2am! It taught us that you don't always need expensive cloud APIs; sometimes the best solution is right on your machine. Our database started simple but got messy fast. We learned that when you delete a user, you really need to think about what else should disappear with them. The resume upload feature was an adventure, turns out converting PDFs to text and storing them is trickier than it sounds.

What's next for ConnectEd

Our roadmap includes advanced AI features like automatic profile summaries, skill extraction from resumes, and smart recommendations based on career trajectory analysis. We'll build an event system with virtual coffee chats, alumni panels, and networking events. An analytics dashboard will provide students insights on profile views and connection rates with personalized improvement suggestions. Mobile iOS and Android apps will enable push notifications and on-the-go networking. Integration with GMU systems will allow profile importing and verified alumni status. Built-in video calling will enable mentorship sessions directly in-platform. A recommendation engine using collaborative filtering will suggest alumni based on similar students' successful connections. Finally, gamification with badges, leaderboards, and rewards will increase alumni participation and student engagement, creating a vibrant, self-sustaining community that makes ConnectEd the default way college students connect with mentors and build professional networks.

Built With

  • anthropic-claude-api-(fallback)
  • axios-(http-client)-###-*backend*-python
  • fastapi
  • github
  • jwt-(authentication)
  • jwt-based-authentication
  • multi-provider-ai-fallback-chain
  • nginx-(reverse-proxy-and-static-file-serving)-###-*development-tools*-git
  • node.js
  • npm
  • ollama
  • ollama-(local-inference)-###-*deployment*-aws-ec2-(t2.micro-free-tier)
  • pip
  • pypdf2-(pdf-parsing)
  • python
  • python-docx-(docx-parsing)-###-*database*-sqlite-with-hybrid-schema-(sql-columns-+-json-blobs-for-flexible-profile-data)-###-*ai/ml*-groq-api-(llama-3.3-70b-for-matching-and-semantic-search)
  • react-router
  • sqlalchemy
  • sqlite
  • tailwind-css
  • uvicorn
  • vite
  • vs-code-###-*architecture*-restful-api
  • zustand-(state-management)
Share this project:

Updates

Submission history