BeesHub

Inspiration

We received a school assignment for a dating app. We were excited to try something new, and this Hackathon well timed.

What it does

BeesHub helps student dev teams:

Match by skills/interests so people swarm the right issues. Auto-suggest collaborators based on task tags and commit history. Recommend “date-like” pair sessions (study jams, bug bashes, coffee-and-code) for faster onboarding and better vibes. Reduce blocking with async task splitting, quick polls for subjective choices, and “Ask the Hive” micro-threads. Visualize progress with a live task graph and health indicators: $$Swarm Health=Active ContributorsOpen Tasks×Resolution Rate\text{Swarm Health} = \frac{\text{Active Contributors}}{\text{Open Tasks}} \times \text{Resolution Rate}Swarm Health=Open TasksActive Contributors​×Resolution Rate$$

How we built it

Architecture

Frontend: React + Vite for snappy DX. Backend: Node.js (Express) with a modular service layer (matching, recommendations, notifications). Data: PostgreSQL (core entities). Collab & DevOps: GitHub (issues, PRs, projects).

Key Features

Skill-Based Matching: Weighted cosine similarity on skill vectors: $$sim(u⃗,t⃗)=u⃗⋅t⃗∥u⃗∥∥t⃗∥\text{sim}(\vec{u}, \vec{t}) = \frac{\vec{u} \cdot \vec{t}}{|\vec{u}| |\vec{t}|}sim(u,t)=∥u∥∥t∥u⋅t​$$ Date Activity Recommendations: Simple rules + collaborative filters from tag co-occurrence (e.g., “frontend + UX” → “Design critique coffee chat”). Non-Blocking Workflows: Automatic task decomposition (labels: good-first-issue, blocked, spike) and async decision polls. Real-Time Hive Chat: WebSockets channels per task with “neighbor ping” to nearby teams for quick help.

Challenges we ran into

Git mishaps: We overwrote work by mistaking which branch was to push or pull. Subjective deadlocks: Debates over display formatting stalled progress. We added time-boxed polls and a “two-way door” rule: reversible decisions don’t need consensus. Coordination friction: Standups turned into blockers. We split work into non-blocking slices and relied on async updates with clear ownership tags.

Accomplishments that we're proud of

Shipped a working MVP with real-time matching, activity recommendations, and a clean task graph. Turned subjective debates into quick evidence-based decisions. Built a resilient-ish Git workflow (branch protection).

What we learned

Avoid arguing subjective features: Use polls, defaults, and “reversible decision” rules. Divide work for non-blocking progress: Slice features vertically; define clear interfaces. How to use GitHub effectively: Issues, labels, PR templates, code owners, and protected branches. Hard lesson: We repeatedly overwrote work—now we treat main as read-only and document recovery steps. Networking matters: Chatting with neighbors saved hours; asking early beats refactoring late.

What's next for BeesHub

Smarter recommendations: Learn from task outcomes to refine collaborator and activity suggestions. Guided onboarding: Interactive “first swarm” walkthrough and environment diagnostics. Notifications for users: Managing 'pending' and 'confirmed' dates and activities. Privacy & ethics: Transparent recommendation explanations and opt-in data controls. Metrics dashboard: Team health KPIs, bottleneck alerts, and trend lines over sprints.

Built With

• express.js
• googlegenerativeai
• node.js
• react
• sqlite
• typescript
• visual-studio-code