Live.ly

Connecting those in need with essential resources in Toronto. Find nearby shelters, food banks, and other resourced using intelligent, compassionate technology tailored to your needs.

Comment

Inspiration

As frequent users of public transit systems like the TTC’s Line 1, our team has witnessed first-hand the increasing number of unhoused individuals navigating life without access to essential resources. It's deeply troubling that, in a city as developed as Toronto, basic necessities remain out of reach for many. Access to shelter, food, and support shouldn't be a luxury– it should be a given.

We set out to build a solution that empowers those currently facing instability, combining modern AI systems and public data to enhance existing resource platforms in a simple, intuitive, and meaningful way.

What It Does

Live.ly offers two core user experiences:

1. For Individuals in Need

Users can make organic, natural language requests like:

“I need a place to sleep tonight. I also need food for the morning.”

Our AI system parses the intent and asks clarifying follow-up questions to gather relevant context (e.g., urgency, age, gender). From there, it intelligently queries public APIs and databases– factoring in shelter availability, proximity, eligibility requirements, and more. These results are then ranked and filtered using custom heuristics and LLM-powered evaluations.

2. For Donors and Organizations

In a future implementation, organizations can bulk-upload their inventory (e.g., food, hygiene products, naloxone kits). Our backend uses AI-based parsing to standardize and store this data in a centralized database. When a user requests specific items, we match their needs with relevant organizations and direct them to pickup locations.

Another future goal is that Live.ly will support verified individual donors, enabling the wider community to contribute safely and effectively.

How We Built It

The backbone of Live.ly is a custom architecture powered by Vellum, which was continually developed and iterated throughout the 36-hour hackathon.

We used:

  • LLM prompt chains and memory via Vellum
  • MongoDB for structured inventory and location data
  • OpenAI for semantic understanding and ranking logic
  • Google Maps API for live shelter and resource location mapping
  • Custom-built web scrapers for food bank and shelter datasets
  • Next.js for the user-facing dashboard

Our team members handled parallel workstreams for UI, backend, and LLM logic to enable rapid prototyping and seamless integration.

Challenges We Ran Into

  • Tool Learning Curve: Adopting tools like Vellum on-the-fly meant we spent significant time learning and iterating workflows.
  • Data Cleaning: Much of the relevant data, particularly for food banks, was unstructured. We developed custom web scrapers and experimented with MongoDB configurations to store and serve this data efficiently.
  • Prompt Optimization: Ensuring the AI could handle edge cases in user queries required multiple rounds of prompt tuning, testing, and fallback chaining. This was made easy through Vellum’s streamlined workflow for rapid iteration and implementing test cases. Features with seeing where certain prompts were able to bypass safety checks were also used to ensure an ideal set of prompts were used.

Accomplishments We're Proud Of

  • Creating a working full-stack platform in under 36 hours
  • Successfully integrating multiple APIs, LLMs, and custom logic
  • Delivering a natural user experience with real-time resource mapping
  • Building something that could genuinely impact lives in underserved communities

What We Learned

  • How to build, structure, and refine LLM-driven workflows
  • The importance of clean, reliable data pipelines in AI-heavy apps
  • Collaborative full-stack development under tight time constraints
  • Real-world use of prompt chaining, memory, and contextual rerouting

What's Next for Live.ly

  • Implementing organizational and individual donors with secure verification workflows
  • Expanding support for multilingual user input
  • Adding further mobile support and offline-friendly features
  • Partnering with local shelters and nonprofits to pilot the platform
  • Further exploration into AI reasoning, such as route optimization and context-aware suggestions

Built With

Share this project:

Updates