TetherAI

Logo
Training results
Training results
User dashboard
Interactive map

Inspiration

Homeless shelters often operate under intense resource constraints, with unpredictable visitation spikes driven by weather, time of year, and external events. After researching the market and getting testimonies from real shelters, we realized that the problems they face can often be much more challenging than we first thought due to the dynamic nature of the issue. Most lack tools to anticipate occupancy or make data-informed decisions. We wanted to create a platform that empowered them with predictive AI, live analytics, and actionable insights, all in a centralized and intuitive webapp.

What it does

TetherAI is an intelligent shelter operations platform built for real-world impact.

Predicts future occupancy using LSTM models with weather and seasonality data
Recommends staffing and supply adjustments with AI-generated suggestions
Visualizes shelter trends on an interactive dashboard calendar
Alerts users to high-risk periods, with explainable reasoning and action items
Supports user authentication, personalized dashboards, and real-time updates

It helps shelter managers anticipate demand and take proactive action instead of reacting too late.

How we built it

Frontend

React.js with TailwindCSS
Custom dashboard cards, calendar visualizations, and map overlays

Backend

Node.js (Express) with MongoDB
REST API with JWT authentication

ML Pipeline

TensorFlow LSTM-based occupancy prediction
Weather + temporal feature engineering with pandas and NumPy
API exposed via FastAPI and served through backend

Other Integrations

Weather APIs
Geolocation sector mapping for generalization
Render for full deployment

Challenges we ran into

Designing a generalizable model that works even for shelters without historical data
Ensuring ML interpretability, turning raw predictions into useful human insights
Combining two tech stacks (Node.js + Python) into one seamless platform
Visualizing time series data clearly across multiple shelters and timelines

Accomplishments that we're proud of

Built an end-to-end platform in under 36 hours with real ML at its core
Achieved high-accuracy occupancy predictions with confidence intervals
Designed a fully modular system with clean code, structured API routes, and multi-agent logic
Created a polished, responsive UI that feels like a production product

What we learned

How to build an LSTM time series forecaster with non-numeric (weather/seasonal) features
How to apply agentic AI patterns to real-world operations workflows
How to balance frontend polish with backend intelligence
How to deliver impact to under-resourced sectors using advanced tools

What's next for TetherAI

Integrating more real-time data from city APIs and service alerts
Expanding support to other cities with plug-and-play geosector models
Partnering with shelters to test real-world efficacy and fine-tune recommendations
Adding multilingual support and low-connectivity fallback modes

TetherAI is more than a dashboard, it's an intelligent ally for shelters that need foresight, not just oversight.