Inspiration
I have watched the field of AI evolve from deep learning models that were being used for text summarization, object notation, semantic search, sentiment analysis among other use cases. These small usecases laid foundations to chatbots that were rule-based and integrated to systems but couldn't provide the natural human to human interaction. At the moment current models are able to naturally interact and generate text in a human-like manner and further more are now able to reason and be integrated with systems giving them the extra ability to perform actions on behalf of users. My aim is to build the best Agentic platform that will enable anyone build and deploy their own AI agents for their own use cases and make existing software applications reason and act through integrations with Agents ushering in the new age of smart software applications.
What it does
Edison, is a platform that let's anyone build and deploy their own AI agents on their website, the web, Facebook Messenger and Whatsapp. You can be able to add tools for your agent to use and also we have our topics feature that enables a user to upload information that their agent will use to answer user queries.
How we built it
Edison is built using Streamlit for frontend, and the backed is built using FastApi. Python is the main language used to build the platform as viewed. And we have used Snowflake cortex to support our RAG tool that every agent will have to help in ingestion and retrieval of contextual information that will be use d by agents to answer user queries. Furthermore, I have also used Deepset haystack to support the agentic AI features and further tools integrations.
Challenges we ran into
Some of the challenges I ran into were really interesting as they offered an opportunity to further improve the platform. They included:
- Ensuring that extracted information can contextually answer user queries.
- Ensuring that context length was not exceeded during user interactions as the retrieved information might be too dense in some cases for the model to generate a proper response and also interrupted user interactions.
Accomplishments that we're proud of
Some of the accomplishments I am proud of include:
- Being able to develop fully the platform and have all the necessary Agentic AI features available on the platform for anyone to use for building and deploying an agent as quickly as possible.
- Having to use Snowflake as the default RAG tool for any agent created to ensure that AI agents.
- Being able to server this platform for other developers and any interested person to use.
What's next for Edison
The next step for Edison is to enable the AI agents to be voice-based and also adding on default platform integrations for users to just easily plug and play. Furthermore, I will intend to expand my existing quota for snowflake service to ensure that users can be able to easily upload content without any service interruption and the agents deployed can be able to perform RAG faster and without any hiccups. Basically, continue building and expanding the platform
Log in or sign up for Devpost to join the conversation.