Built at DemonHacks 2026 — DePaul University's UPE & CSS Hackathon
An AI-powered desk-sharing marketplace that turns empty corporate offices into startup launchpads.
Chicago's Loop is the economic heart of the city — but since the shift to remote work, office occupancy has plummeted. The ripple effect is devastating: cafeterias, restaurants, and small shops that depend on foot traffic are closing one by one. Entire blocks are going dark.
LoopShare is our answer. We connect corporations sitting on empty floors with startups hungry for affordable, flexible desk space in a premier downtown location. More desks filled means more workers in the Loop, more lunch orders, more coffee runs, more life on the streets — revitalizing the small businesses that make the neighborhood thrive.
LoopShare serves two types of users:
Companies with underutilized office space in the Loop. They can:
- Post a Listing — List available desks with pricing, days, floor, and amenities
- Tax Estimator — Calculate Illinois Enterprise Zone Act tax savings from desk-sharing revenue
- Host Dashboard — Manage listings, view incoming bookings, and track occupancy
Small teams looking for flexible, affordable desk space. They can:
- Find a Host — Browse all active listings with filters, maps, and building details
- AI Match — Fill out a profile (sector, team size, budget, preferred zone, days needed) and let our AI pipeline find the top 3 best-fit offices in seconds
- Find New Host — Our ML model proactively identifies "ghost buildings" with declining occupancy — buildings ripe for desk-sharing before they even list on the platform
- Startup Dashboard — View bookings and saved listings
The platform is built from four modules:
A single-page app with interactive Leaflet maps, Chart.js visualizations, and a clean onboarding flow. The Vite dev server proxies API calls to both the backend and the ML service.
Tech: React 19, React Router 7, Leaflet, Chart.js, plain CSS
The REST API layer handling users, listings, bookings, buildings, and the AI orchestration pipeline. Uses an H2 in-memory database seeded with 21 real Loop buildings, 8 hosts, 10 listings, and 10 startups.
Tech: Java 11, Spring Boot 2.7, Spring Data JPA, H2, Swagger UI
A Gradient Boosting Regressor trained by us on real Chicago Energy Benchmarking data. The model uses building energy usage (EUI, electricity, natural gas) and physical attributes (floors, year built, square footage) to predict occupancy rates. Buildings where predicted occupancy has dropped significantly since 2019 are flagged as "ghost buildings" — prime candidates for desk-sharing outreach.
Tech: Python, scikit-learn, Flask, pandas — trained on 10+ years of public Chicago energy data
A multi-agent system powered by a locally-running Ollama LLM — no paid API keys, no cloud dependency. The four agents work together in a single orchestrated pipeline:
| Agent | What it does |
|---|---|
| MatcherAgent | Scores listings against startup needs using a deterministic formula + real energy data, then asks the LLM to explain each match in natural language |
| AnalystAgent | Tool-use agent that autonomously calls Chicago public APIs (energy, violations, CTA ridership) to produce a deep building intelligence report |
| RiskAgent | Scores deal risk (0–100) for both host and tenant by checking business licenses and building violations via the LLM |
| OutreachAgent | Generates a personalized outreach email and a one-page lease draft using all data gathered by the other agents |
The LLM is also used in Deal Scout to enrich company profiles for each building and draft outreach messaging — all running locally on llama3.2 with zero API costs.
# 1. Backend
cd demo
./mvnw spring-boot:run # Windows: mvnw.cmd spring-boot:run
# Runs on http://localhost:8003
# 2. ML Model (new terminal)
cd occupancy_model
pip install -r requirements.txt
python api.py
# Runs on http://localhost:5000
# 3. Frontend (new terminal)
cd frontend
npm install && npm run dev
# Runs on http://localhost:3000
# 4. Make sure Ollama is running with:
ollama pull llama3.2
ollama serve| Prerequisite | Version |
|---|---|
| Java | 11+ |
| Node.js | 18+ |
| Python | 3.9+ |
| Ollama | latest (ollama.com) |
No Docker, external database, or cloud API keys required. Everything runs locally.
|
Every data point in LoopShare comes from the Chicago Open Data Portal — free, open, and maintained by the City of Chicago. No proprietary datasets, no scraping, no API keys. |
|
| Dataset | What we use it for |
|---|---|
| Energy Benchmarking | EUI scores, ML model training, building efficiency |
| Building Violations | Risk scoring (fire, structural, electrical) |
| CTA Ridership | Occupancy proxy for the Loop district |
| Business Licenses | Company verification and risk assessment |
Because LoopShare is built entirely on public data, it can serve as a tool for City Hall and the State of Illinois — not just startups and corporations. The city already collects this data; we turn it into actionable intelligence:
- Chicago City Hall could use our ghost-building detector to identify underutilized properties for economic development programs, zoning incentives, or targeted outreach.
- The State of Illinois could track Enterprise Zone Act utilization and measure desk-sharing's impact on downtown economic recovery.
- Small business coalitions could use Loop occupancy trends to advocate for foot-traffic revitalization programs.
Open data in, public good out. That's the idea.
- Diego Fernandez Arias
- Laura Rueda Garcia
- Pablo Bote Lopez
- Hernan Garcia Quijano
Built at DemonHacks 2026 by the LoopShare team. Revitalizing Chicago's Loop — one desk at a time.
