TeamSync AI: Eliminating the "Tragedy of the Commons" in Group Work
Inspiration: Every hackathon and team sprint faces the same hidden bottleneck: coordination failure. When projects start, tasks are split haphazardly in chat threads or spreadsheets. One person ends up working 18 hours on backend and frontend while another is underutilized or stuck on mismatched tasks. We wanted to eliminate this "Tragedy of the Commons" in group work by building TeamSync AI—an intelligent workload coordinator that uses mathematical fairness and open-source LLMs to distribute tasks with absolute transparency.
What it Does: • 1-Click Hackathon Setup: Leads can load pre-configured hackathon rosters and deliverables instantly, or input custom members, skills, and hours. • Fairness Scoring Kernel: Evaluates assignments against skill compatibility and hour variance, computing a real-time Workload Fairness Score. • Transparent AI Reasoning: Every task card displays the assigned member alongside a 1-sentence explanation of why the assignment was made. • Worker-Specific Views: Individual teammates can filter the board to "My Tasks" to see their exact commitments without dashboard clutter. • Secret Judge Tour (/judge): An Instagram-style scrollytelling walkthrough detailing our mathematical formulation, security layer, and architecture.
How We Built It: • Backend: Built using FastAPI and served with Uvicorn, handling task coordination, role sessions, and static assets. • AI Orchestration: Integrated via OpenRouter API with qwen/qwen-2.5-coder-32b-instruct as primary, meta-llama/llama-3.3-70b-instruct as fallback, and a built-in offline mock engine. • Reliable Extraction: A 3-stage regex extraction pipeline parses structured JSON payloads out of model outputs reliably. • Security & Sanitization: Custom security middleware (security.py) enforces strict boundaries (max 10 members, max 20 tasks), neutralizes prompt injection markers, and escapes HTML characters against XSS. • Frontend & Kinetic UI: Crafted with vanilla HTML5, CSS3, and modern JavaScript using our signature palette: Raw Ochre (#c25e00), Weathered Slate (#0f172a), and Industrial Cyan (#00f0ff).
Mathematical Fairness Kernel: To avoid arbitrary distribution, our scoring kernel balances skill matching against the normalized variance of assigned hours:
(\text{Workload\ Fairness\ Score}=100\times \left(1-\frac{\sigma _{\text{hours}}}{\mu _{\text{hours}}}\right)\times \text{SkillMatchFactor}) Where: • (\sigma _{\text{hours}}) is the standard deviation of assigned hours across teammates. • (\mu _{\text{hours}}) is the mean assigned hours. • (\text{SkillMatchFactor} \in [0.0, 1.0]) represents how closely member proficiencies align with task requirements.
Challenges We Faced:
- Model Output Reliability: Small open-source models occasionally wrap JSON in explanatory text or irregular Markdown blocks. We engineered a multi-stage regex fallback parser to guarantee zero frontend crashes.
- Preventing API Lockouts: During heavy hackathon testing, rate limits can block live demos. We built a zero-latency local fallback engine that automatically engages if keys are exhausted.
- Designing for Scrollytelling: Implementing sticky scroll transitions and animations in pure vanilla CSS and JavaScript without heavy external libraries required precise tuning with Intersection Observer.
What We Learned: • How to structure production-grade fallback chains across open-source LLMs. • Implementing defensive prompt injection and bounds sanitization directly in API pipelines. • How to craft high-conversion kinetic micro-interactions that communicate mathematical proofs clearly to end users.
What's Next for TeamSync AI: • Direct bi-directional integration with GitHub Issues, Jira, and Slack. • Multi-sprint dependency scheduling to handle sequential blocking tasks automatically. • Team burnout alerts that flag sustained high-effort sprints before deadlines hit.
Built With
- css3
- fastapi
- html5
- javascript
- llama-3.3
- openrouter-api
- pydantic
- python
- qwen-2.5-coder
- rest-api
- uvicorn
Log in or sign up for Devpost to join the conversation.