Inspiration
If you've ever assembled a 3D printer or any other product that requires extensive consultation with manuals, along with countless online searches to find answers to questions the manual doesn't address, you know the frustration I'm talking about. My idea is to build an agent that simplifies the installation process for any complex item.
What it does
The agent stores the state of the installation the user is at at any moment, and via input from the user updates the state the user is at. The user can query any question from the agent. The agent will try to find the answer from the manual and/or from online searches.
Challenges we ran into
- Dynamically defining the state of the agent when consuming a new manual
- Building RAG for large documents
Accomplishments that we're proud of
- Making the pipeline almost functional
What we learned
My main takeaway is the feasibility of building this product and its high usability potential. I believe the quality of my agent is closely tied to the RAG model, and with improvement in newer models as well as the search engine tool, the quality of the agent's outputs will also be improved.
What's next for AssemBuddy
Introducing multimodal Retrieval-Augmented Generation (RAG) and image search to the agent's answering system
Built With
- langgraph
- openai
- python
Log in or sign up for Devpost to join the conversation.