Alms | Devpost

For Evaluators/Sponsors

Scroll down for a handy guide to navigating our repository and our project's assets.

💥 - How it all started

In San Francisco, a city known for its vibrant culture and bustling streets, there lies a less visible yet significant issue: vehicular homelessness. Over 2,000 individuals call their vehicles home, facing daily challenges that most city dwellers rarely consider. This reality hit close to home for our team, as we've encountered friends and acquaintances grappling with finding safe and legal parking spaces, leading to heightened stress and uncertainty. These experiences ignited our passion for devising a solution to alleviate the struggles faced by those living in their vehicles.

The problems encountered by the vehicular homeless community are multifaceted and complex:

Finding Safe Parking: Locating a safe and legal parking spot for the night is a daily struggle, with many areas imposing restrictions on overnight parking.
Community Disconnection: Those living in their vehicles often feel isolated and disconnected from community resources and support networks.
Legal Challenges: The constant threat of fines, towing, and legal repercussions for parking in unauthorized areas exacerbates the stress of vehicular living.
Access to Basic Amenities: Finding access to basic amenities such as restrooms, showers, and laundry facilities while living on the move is a constant challenge.

📖 - What it does

To address these challenges, we developed Alms, an innovative app designed to connect individuals living in their vehicles with available parking spaces. Alms is the culmination of cutting-edge technology and compassionate community outreach, engineered to make vehicular living safer and more manageable.

Here’s how Alms leverages advanced technologies to serve its mission:

Safe Parking Availability: Utilizing React for its user-friendly interface, Alms provides a seamless experience for users to navigate through a real-time map of available parking spots. These spots are designated for overnight stays, including those offered by supportive community members, churches, and businesses.
Building Community Connections: The app encourages local communities to engage with and support their vehicular residents. This is facilitated through FastAPI, ensuring robust and scalable back-end services that handle user interactions efficiently, fostering a sense of community and belonging.
Legal Peace of Mind and Access to Amenities: By integrating Neurelo for intelligent recommendations and Fireworks.ai for predictive analytics, Alms is able to anticipate the needs of its users, directing them to legally safe parking spots and nearby amenities such as showers, restrooms, and laundry services. This not only reduces the legal risks associated with unauthorized parking but also ensures users have access to essential services.
Enhanced Communication and Support: Leveraging FireFunction V1's function calling and Mixtral MoE's intelligence, Alms provides an empathetic and responsive AI assistant that offers support, answers questions, and guides users through the app. This AI-driven approach ensures users have instant access to the information they need, enhancing the overall user experience and providing emotional support.

Alms is more than just an app; it's a movement towards creating a more inclusive and supportive environment for those living in their vehicles. By combining the power of community support with advanced technology such as React, FastAPI, Neurelo, and, Fireworks.ai, we aim to bridge the gap between the housed and vehicular residents, ensuring that everyone in San Francisco has a safe place to rest at night.

🔧 - How we built it

To address the challenges of vehicular homelessness with our app, Alms, we've woven together a fabric of cutting-edge software technologies, each selected for its ability to enhance the app's functionality and user experience. Here's how each technology contributes to Alms:

Technologies

React: Powers the user interface, offering a seamless and responsive experience that makes it easy for users to find available parking spots and connect with community resources.
FastAPI: Provides the backbone for the backend, handling requests efficiently and ensuring scalability to support a growing user base.
Mapbox: Integrates detailed, interactive maps into Alms, allowing users to visually navigate and locate safe parking spots in real-time. This geospatial data visualization tool is crucial for users to find and understand the availability of parking spaces and community-offered amenities.
Neurelo: Used to provide a cloud data api with an intuitive web UI to generate, test, ad troubleshoot our data queries and API.
Fireworks.ai: Hosted the language models (FireFunction V1 and Mixtral Instruct) that power our AI chatbot. Enhanced Communication and Support

The integration of these technologies into Alms creates a comprehensive solution to the complex issue of vehicular homelessness. React and FastAPI ensure a robust and efficient app infrastructure, while Mapbox brings an essential layer of geospatial data visualization, making the search for parking intuitive and user-friendly. Neurelo and Fireworks.ai streamlined and enhanced our software development process. Lastly, the inclusion of FireFunction V1 and Mixtral Instruct offers a responsive and understanding AI assistant, making Alms not just a tool, but a supportive companion for those in need. This synergy of technologies makes Alms a powerful platform for individuals living in their vehicles, providing not only practical solutions for parking and amenities but also fostering a sense of community and belonging in San Francisco.

🚩 - Challenges we ran into

Neurelo and PostgreSQL Integration: The primary challenge was to effectively use Neurelo with our PostgreSQL database, which contained complex geometric data types that Neurelo wasn't inherently designed to handle. The workaround involved extensive collaboration with George, an expert on Neurelo's querying capabilities. By devising custom query functions, we navigated around the limitations of geometric data. This led to the creation of large SQL queries, which, thanks to Neurelo's robust abstraction, became seamlessly manageable within our application.
Innovative Use of Function Calling with LLMs: A notable learning curve was encountered when attempting to employ a standard chat/completion functionality of a large language model (LLM) for multiple distinct tasks. The breakthrough came from Ray Thai at Fireworks.ai, who introduced the concept of function calling within LLMs. This technique allows the model to be supplied with an assortment of "tools," enabling it to select and execute tasks more efficiently, thus broadening the LLM's utility within our project.
Challenges with Reload Function Implementation: Ensuring a smooth and responsive user experience required the development of a reliable reload function. The difficulty lay in creating a function that could consistently update and display the latest data without glitches or delays. This function was critical for real-time updates, such as the availability of parking spots or amenities like restrooms and Narcan, and necessitated meticulous testing and refinement to meet our performance standards.

🏆 - Accomplishments that we're proud of

Going from idea generation to a user-oriented and responsively made application that is not far from public deployment
Integrating machine learning models and large language models within our application in a way that is directly accessible to users
Being adaptive to feedback from sponsors and incredible mentors like Jacobson and Vinny
Working seamlessly as a team and take advantage of sponsored tools like Fireworks.ai and Neurelo to make our vision into a reality

📝 - What we learned

Ameya: "Learning about the way we can leverage Fireworks.ai to engineer our machine learning assistant for Alms was really cool."
Connie: "I learned about JavaScript. I learned about prompts and React."
Nicholas: "I learned about function calling for LLMs from Fireworks.ai's Ray Thai, which enables multi-tasking within a single model."
Alex: "Mastering Neurelo to interface with PostgreSQL's geometric data was challenging, but with George's guidance on custom queries, we significantly streamlined our database interactions."

✈️ - What's next for Alms

Just like how a great trip has a great itinenary, we envision Alms future plans in phases.

Phase 1: Solidifying our Roots

Phase 1 involves polishing our user interface and creating an app version of our mobile experience.

Phase 2: Branching Out

Phase 2 involves creating a ranking system that takes in all parameters and provides a desirability score and expanding the information provided and analyzed in Alms.

Phase 3: Take Off

Phase 3 involves expanding accessibility of the app through having our services be available in numerous different languages and establishing a pilot program in SF.

📋 - Evaluator's Guide to Alms

Intended for judges, however the viewing public is welcome to take a look.

Hey! We wanted to make this guide in order to help provide you further information on our implementations of certain programs and provide a more in-depth look to cater to both the viewing audience and evaluators like yourself.

Sponsor Services We Have Used This Hackathon

Fireworks AI

Used the FireFunction V1 and Mistral MoE 8x7B Instruct models to enable our chatbot. The way our chatbot works is it first uses Mistral to determine what service the user requires given their messages. If the service required is parking, it uses the FireFunction V1 models to generate clarifying questions and gather the necessary data to help users find parking. If the user requires another service, we fetch the most relevant resources and provide the user with the ones closest to them.

Neurelo

By using Neurelo as cloud api between our server and database, we were able to quickly generate, test, and manage complex data queries. Also, by developing our data api through Neurelo, we were able to use the intuitive web ui to work on API development rather than in raw, cluttered code.