DeepDive

• 

  Interface

Inspiration

While researching potential issues to take on, we came across the uniquely challenging profession of underwater welding. With a mortality rate as high as 15%, underwater welders experience monumental risk despite being crucial to the operation of various global infrastructures. In an effort to reduce potential harm due to mismanagement, we sought to create a system that would streamline job assignments while taking workers' safety and interests into consideration.

What it does

DeepDive is a workforce management system that utilizes AI reasoning to recommend personnel for job assignments. The program allows users to track company employees, along with their capabilities, limitations, and availability. Users can enter descriptions of new jobs to receive AI recommendations to suggest the most suitable workers according to the relevant provided information and standard regulations to be followed.

How we built it

Our application is built using Python, LangChain, MySQL, and React, following a modular multi-tier architecture. The core-service acts as the central data layer, managing our MySQL database and determining what information to pass to our AI agents. The agentic-service sits between the core and the agents themselves, orchestrating agent execution, providing deterministic tools, and maintaining a RAG pipeline. Finally, our React frontend ties everything together, offering a clean interface for interacting with the system.

Challenges, Accomplishments, Learning Outcomes

Despite our relative inexperience with hackathons and some initial trouble integrating Gemma, we were able to leverage new technologies and enjoy our experience over the last 36 hours. During our time at HackTech, we identified a problem space and produced a system that could be used to address it. By creating this project, we were able to go through the steps of the full lifecycle of a start-up company, from ideation to development, and finally, presentation of the proof of concept.

What's next for DeepDive

From developing our initial system, it became evident that many improvements were possible for future iterations. The next milestone for DeepDive would be refining our systems to be able to address a wider problem space in various professional settings. Further optimization would include switching towards a multi-agent pipeline to improve AI response time while maintaining quality and potentially adding features that would let our highly structured employee definitions be automatically inputted from other management software.

Built With

• fastapi
• gemma
• javascript
• langchain
• mysql
• python
• react