Inspiration
With the recent advancements in generative AI we have seen a rise in the capability of LLMs to reason and perform tasks. This is a huge step in making AI practical and applicable, by integrating it to various systems to help in automation of various business processes and produce better outcomes for customers and businsses.
What it does
Edison Functions, is a platform that enables businesses or anyone to explore and deploy AI agents. Through Edison Functions, you can create an agent, integrate it with a system by adding tools it can use and deploy it for interaction with consumers on Facebook Messnger and WhatsApp.
How we built it
First, the backend service first was developed to enable user onboarding and add functionality to create the agents. Secondly, we built the generator module which houses the internal logic of how agents will work. Thirdly, we focused on providing webhook functionality that will be used to provide the agent with the ability to execute actions on behalf of the user.
Challenges we ran into
- Agent response accuracy - The LLM being used, Meta-Llama-3.1-70B-Instruct, has not been properly fine-tuned for tool use.
- Out-of-the-box integrations for the agents - Currently we are working to develop out of the box integrations that will make it easy to integrate an agent. At the moment the focus is on providing out-of-the-box online retail stores specifically through the platforms Woocommerce, Shopify and Magento.
- Serving the agents - Currently, agents can be deployed on Facebook Messenger and WhatsApp. We are working to design an additional interface that can be used to provide access to the built agents via the web for more complex use-cases through generative UI.
Accomplishments that we're proud of
- Quick design and deployment of agents - Through Edison Functions, anyone can develop an agent and deploy it for user interaction on social media.
- Actions to boost agent functionality - Through Edison Functions, developers can add various actions that an agent can perform on behalf of users.
- Agent response speed - Through the use of the models served by SambaNova agents responses are very fast, which is important when an agent is executing an action and conveying the result of the action to a user.
- Being involved in the revolution to bring Agentic AI to the masses - We look to make Edison Functions a platform that democratizes AI agents to everyone and the leading platform for quickly configuring, testing and deploying Agentic AI use-cases.
What we learned
- Platform improvement - We are aiming to provide multi-agent functionality on the platform, this will enable our users to spring up multiple agents that can work together.
- Agentic AI is the future, we need to add more in-built integrations that created agents can be easily connected to increase use-cases to AI agents.
What's next for Edison Functions
The next step is to provide custom integrations enabling businesses to build and deploy their AI agents quickly. Currently, our focus is on online retail stores, enabling them to easily integrate their stores with the platform and deploy AI agents that help them explore conversational commerce. The next focus on the healthcare sector through providing integrations to systems like OpenMRS and DHIS2.
Log in or sign up for Devpost to join the conversation.