Radar

Setup complex alerts
Get recommended alerts
Pause Alerts
Notifications
Manage alerts via UI

Inspiration

A lot of data is captured in Salesforce

As a developer supporting systems like Salesforce, we constantly get requests to create reports and notifications
When we create these reports, there is always a bunch of follow-up requests to make updates, and some teams shift what they look at quite often
Do sales teams, customer support teams, finance teams, and XIRs need all these reports? A lot of times, teams just want to know when certain things do (or don't) occur so that they can respond accordingly

What it does

Radar allows users to say what they're interested in monitoring using natural language, and AI is used to interpret what needs to be monitored.

You can create alerts using natural language: "Let me know when Acme creates a new issue."
- Pause alerts "Mute alerts for this week, I'm on vacation"
Get recommended alerts based on your goals, "we're trying to increase revenue. What should we keep an eye on?"

Alerts are triggered when records are updated or on a schedule, and a notification is sent to the user.

How we built it

Radar is built using Salesforce's Agentforce framework to handle natural language requests. We used actions to create and evaluate alerts, while Flows and Prompts orchestrate the notifications users receive.

This architecture lets us combine flexible AI-driven interactions with deterministic Salesforce processes.

Challenges we ran into

One challenge was figuring out what an action within a topic should be instead of an action within a Flow. Eventually, I realized that where Agentforce shines is orchestration actions, but if you want actions to execute in a deterministic way, you should use Flows.

Accomplishments that we're proud of

Getting the LLM to handle a broad range of alerts
Recommended alerts and getting the LLM to probe deeper so that recommendations are relevant

What we learned

Learned a lot about the Salesforce ecosystem and how easy it is to build applications for the platform.
Agentic apps are more fault-tolerant than traditional applications. Typically, your application requires specific information to function. Still, with Agentforce, you can get the best of both worlds, where you define what is needed but use instructions to have the LLM work towards getting the required info. If things are misspelled, it still understands
Marrying a chat experience with a traditional UI offers significant advantages