As you know, Glowstick detects sales opportunities in customer meetings. For each opportunity detected, we extract the content and generate a headline using OpenAI GPT-4.
We recognized that our headline generation has been subpar for a while and we wanted to change that! We organized a power-hour session with the entire company to collectively tackle the problem and find a solution!
During the workshop, every team member was encouraged to create their own agent to contribute to the improvement of headline generation. This approach allowed us to leverage the diverse perspectives and expertise within our team. During the power session, everyone was invited to share what worked and what didn’t. By doing this, we quickly learned how to craft a prompt that would suit our needs.
Quick schedule for an hour session.
- Invite all team members to your OpenAI Organization and send them our Loom for explanations!
- Create a dataset with 4-5 examples from real data. See data.json to see how we do it.
On the day of the workshop
- Take 2-3 minutes to go over the rules again and ensure that everyone can use the OpenAI Playground.
- First Round: 10 minutes.
- People individually try prompts on the dataset.
- Sharing: 15 minutes.
- Roundtable where we can share our prompts, what worked, what didn’t, and what we learned.
- Second round: 10 minutes.
- Get the best prompt possible and send your assistant id to the owner!
After the workshop
- Use the code provided to generate summaries for each agent and each test sample.
- Ask people which summary is the best for each test sample.
- The most liked assistant wins!
Our ML team has a Loom and resources on how to start using the OpenAI platform:
Prompt Engineering resources
After careful evaluation, the winning agent was selected and deployed to production! Congratulations to Shruti and their awesome agent for making Glowstick better!
This approach not only improved our headline generation, but also fostered a collaborative environment where everyone had the opportunity to contribute and be part of the solution. This makes the deployment of new features much easier as anyone can now create their agent.
Stay tuned for more updates and insights as we embark on new challenges and innovations in the field of AI!