AgentforceHackathon

• 

  Experience site image

• 

  A prompt that predicts hazard maps based on the addresses of business partners and suggests recommended products for each hazard map

Inspiration

I wanted to make something with Agentforce. And it was something I was most interested in!

What it does

It was created using flows and prompts. Based on the expiration date of the shipping list, it searches for products that are about to expire. The prompt also suggests recommended disaster prevention supplies from the hazard map. The prompt also takes recently arrived products and the customer's hazard map as input information and suggests five recommended products from those. In addition, a flow was created to obtain customer information using an email address as a condition. When placing an order, it is also used to obtain the product ID from the product code and to create order information. A command was entered in the prompt to create it in JSON format.

How we built it

It was created using flows and prompts. Products with an upcoming expiration date are retrieved based on the expiration date in the delivery list. The prompt also presents recommended disaster prevention products from the hazard map, and the prompt uses recently arrived products and the customer's hazard map as input information to present five recommended products from these. I also created a flow that retrieves customer information based on email as a condition.

Challenges we ran into

As expected, we struggled with the design of mapping the hazard map with products and presenting appropriate products. We spent a lot of time narrowing down the products because we thought that it would be too slow to bring in all products unconditionally and have LLM choose from them if there were a lot of products, but we eventually settled on having recently arrived products or something like a recommended product flag to extract products and have LLM choose recommendations from that. Also, when creating order information, we specified the JSON format, passed it directly to Apex, and registered it as an object, but it took some time to modify the prompt and try and error until we found a format that could be mapped to the object.

Accomplishments that we're proud of

Even if you look at a hazard map, it's still hard to decide what you should prepare, so I think there's a lot of demand for products that can be proposed while looking at these hazard maps. Hazard map conditions can change quite a bit, so I think it's a good solution to have LLM propose products with a certain degree of accuracy, rather than having it inside SF.

What we learned

At first, I didn't know what to do and how to do it, but I learned that by breaking down the work little by little and adding actions to Agentforce, it gradually became possible to have a conversation. It was a bit of a trial and error process, but what I could do yesterday might not work today, so I figured out the cause when I couldn't and added it to the instructions. I realized that it's important to test repeatedly.

What's next for AgentforceHackathon

I would like to actually implement Agentforce into our customers' business operations.

Built With

• flow
• salesforce