Armada AI

• 

  Landing Page for Armada AI

Inspiration

After doing lots of customer discovery, we met Dave Brautigan - a Georgia Tech grad, private equity operator, and owner of Rocky Mountain Pizza. In his portfolio, he has many different commercial food service technician companies and he mentioned that his biggest problem is his technicians finding the right industrial parts. These after-market & OEM parts are scattered across the web, spread in various retailers, gate-kept by manufacturers; making this search process very manual for technicians. Armada AI is an artificial intelligence application that allows technicians to chat with it to find the right part in seconds. We are revolutionizing procurement & the search process for technicians across the globe.

What it does

Armada AI is the go-to copilot to assist technicians in procuring the supplies and parts they need for their jobs. By prompting our expert AI procurement agent Andy, service technicians can identify the correct part required for each job and are linked directly to a website to place their order. Armada's MVP focuses on gasket replacement and repair but will expand to a massive library of parts, suppliers, and general repair advice.

How we built it

Armada AI's Andy is a GPT-4 powered assistant. We wrapped GPT-4, aligning their LLM to generate responses based on a document containing information scraped from AllPoints, the largest parts supplier in the industry. We used Selenium to scrape information from AllPoints into a beautiful soup and extracted pertinent information to create a JSON document used for GPT's context. After receiving accurate responses, we designed and developed a chatting ui with a React frontend paired with a FastAPI backend to call the GPT-4 assistant. Finally, we released Andy to the world by hosting our app with Vercel.

Challenges we ran into

One challenge we ran into early on was higher costs than expected for calling GPT-4's API. To reduce costs, we restructured our code to re-use one assistant rather than creating a new assistant every time. Another challenge we faced was version control in GIT. For the react app in particular, all of our team members were running different versions of node which caused frequent breaks. Finally, API key security posed a challenge -- storing our GPT-4 API key as an environment variable in our Vercel build was a mountainous obstacle.

Accomplishments that we're proud of

1) Secured a Letter-of-Intent (LOI) from Diversified Food Services Inc. to provide an AI/high-tech solution for their portfolio companies. 2) Spent 36-hours in an all-out sprint to build a lean MVP that we can pitch to secure potential contracts & clients. 3) Absolutely working our butts off and learning how each of our teammate’s complimentary skills allows us to move together as a unit.

What we learned

1) Currently, the scalability of AI tools is not budget friendly towards smaller teams & developers. The monetization is still being flushed out and builders will have to bear some expense for ‘experimenting’ when trying to create the future. 2) We also found that the actual value output to end users is directly proportional to how clean the data is and how well it’s organized.

What's next for Armada AI

1) Pitching our MVP to Diversified Food Services to secure our first paid customer. 2) Networking with Atlanta VCs & angels to potentially raise capital.

Built With

• fastapi
• openai
• python
• react