jaimy.ai

your workplace wingman

Comment

Inspiration

When you receive a Calendly link, it often feels like an impersonal gesture, lacking the warmth and thoughtfulness that personal interactions demand. This sentiment is shared by thousands of professionals across various industries. In fields like finance, consulting, and law, the personal touch is everything. The way we communicate can make or break relationships, deals, and opportunities. Recognizing this need, we created Jaimy—a revolutionary autonomous agent designed to transform business communication. Jaimy aims to restore the personal touch while enhancing efficiency. It helps professionals excel by making them more effective communicators and ensuring that tasks are managed seamlessly.

What it does

Jaimy is a game-changer in the realm of business communication and task management. In this project demo, Jaimy showcases its ability to autonomously book meetings, whether in-person or virtual. This process is entirely automated—just CC Jaimy on your emails, and it will handle the rest. Jaimy reaches out to all involved parties, finds a suitable time, and schedules the meeting. If the other party also uses Jaimy, the interaction becomes fully automated, leading to a perfectly scheduled meeting in your calendar without any manual intervention. Additionally, Jaimy provides a comprehensive record of the email chain, ensuring transparency and clarity.

Beyond scheduling, Jaimy is adept at managing day-to-day tasks. Have an event you need to add to your calendar? Simply send an email to Jaimy, and it will take care of it. This capability extends to various administrative tasks, making Jaimy an indispensable tool for professionals who need to optimize their time and streamline their workflow. By handling these mundane tasks, Jaimy allows users to focus on more critical aspects of their work, thereby enhancing productivity and efficiency.

How we built it

The development of Jaimy involved integrating multiple advanced technologies to create a seamless user experience. Jaimy consists of three primary components: the Action Classification Layer (ACL), the External Resources Bridge (ERB), and the Core Agent (CA). The ACL leverages sophisticated models through Groq such as Llama-70B to accurately classify the user's desired action from their communication. This classification is critical for ensuring that Jaimy understands and performs the correct tasks.

The CA is the heart of Jaimy, connecting with the ERB to form a continuous feedback loop until the user's task is completed. The ERB integrates various external resources, including Google Calendar, email services, and web surfing capabilities. This integration allows Jaimy to access and utilize these resources efficiently. When tasked with an assignment, the CA processes all relevant information, including the entire email chain, and collaborates with the ERB to make necessary calls and decisions. This architecture ensures that Jaimy operates smoothly and effectively, providing users with reliable and timely support.

Challenges we ran into

Developing Jaimy was not without its challenges. One of the most significant issues we faced was managing hallucinations and inconsistencies in the responses generated by the agents. The internal processes of the agents interacting with each other can be a black box, often leading to random terminations of tasks. This problem was particularly pronounced with long email chains, as the agents needed a perfect sense of logic and continuity to understand the overarching task. To address this, we implemented a system to store message threads in our database, allowing us to select specific ones for the agent to consider. This approach, combined with simplifying the CA, improved the overall efficiency and reliability of Jaimy over time.

Accomplishments that we're proud of

One of our proudest accomplishments is the successful integration of all components into a cohesive and functional system. Initially, we had individual parts working—calendar integration, room booking code, email chain management, and the agents themselves. However, combining these elements into a seamless solution presented numerous challenges. Each component's issues compounded when integrated, requiring continuous prompting and adaptation of the agent with generated content and context.

Through persistence and innovative techniques, such as utilizing enums and leveraging function calls, we overcame these challenges. The result is a working demo that not only meets our initial goals but also exceeds expectations in terms of functionality and user experience. Seeing Jaimy in action, autonomously handling complex tasks and communication, is a testament to our team's hard work and ingenuity.

What we learned

Throughout the development of Jaimy, we learned that integration is the most challenging aspect of building a complex system. While each component may function well individually, integrating them into a seamless whole requires meticulous attention to detail and problem-solving. This experience highlighted the need for continuous testing, refinement, and optimization to ensure that all parts work harmoniously.

We also realized that the landscape of large language models (LLMs) is still evolving, and there is significant potential for improvement. Our experience suggests that focusing on smaller, more specialized models could yield better results than relying solely on large, general-purpose models like GPT and Llama. These specialized models could be fine-tuned to handle specific tasks more effectively, providing more accurate and reliable performance.

What's next for jaimy.ai

Looking ahead, our goal is to expand Jaimy's capabilities to handle all aspects of business communication. We envision Jaimy being able to draft emails, monitor inboxes, proofread documents, book client meetings at restaurants, and even manage traditional communication methods like mailboxes and fax machines. By expanding its functionality, Jaimy will become an even more indispensable tool for professionals across various industries.

Additionally, we plan to develop and fine-tune our own smaller, specialized models. While GPT and Llama have been invaluable, they can be overkill for certain tasks and sometimes lack the necessary precision. By creating more efficient, task-specific models, we aim to enhance Jaimy's performance and reliability further. This approach will allow us to provide a more tailored and effective solution to our users, solidifying Jaimy's position as a leader in business communication and task management.

Share this project:

Updates