Inspiration

Project Report: PulsePoint — The Autonomous Analytics Bridge

Inspiration: The "Static Insight" Problem

The genesis of PulsePoint was born from a common frustration in modern enterprise environments: the "Last Mile" gap between data visualization and business execution. While Tableau provides world-class descriptive analytics, the insights often remain trapped within dashboards, requiring manual human intervention to translate a "red" metric into a Salesforce action. We were inspired by the concept of Autonomous Analytics—a world where the data doesn't just tell you what happened, but actively triggers the systems that can fix it.

Our vision was to bridge Tableau Cloud's analytical depth with the operational power of Salesforce Agentforce. We asked: What if the dashboard could think?

How We Built It

PulsePoint is built on a modern React stack, utilizing the Tableau Developer Platform and Gemini 3 Pro as the cognitive engine. The architecture follows a three-tier model:

Ingestion Layer: We utilize the Tableau Metadata API to crawl workbook structures, identifying critical KPIs like Customer Churn Risk and Revenue Pipeline.
Reasoning Layer: Using the Gemini API, we process this metadata to identify anomalies. We represent the efficiency of our autonomous agent using the following LaTeX notation: $$E_{total} = \sum_{i=1}^{n} (T_{manual} - T_{pulse})i \times C{resource}$$ Where $T_{manual}$ is the time taken for a human to act on an insight, $T_{pulse}$ is the near-zero latency of our agent, and $C_{resource}$ is the operational cost.
Action Layer: This tier connects to Salesforce Data 360 to pull unified customer profiles and triggers Agentforce "Actions"—such as automated case generation or Slack-based emergency alerts.

Challenges Faced

The primary challenge was Context Synchronization. Tableau and Salesforce often speak different "languages" (different UUIDs and object schemas). Mapping a Tableau Viz ID to a Salesforce Account ID required a robust translation layer. Furthermore, we faced significant hurdles in "Thinking Budget" management for the AI; we had to optimize the prompt to ensure the model stayed within the reasoning limits for complex multi-step tasks like: $$P(Action) = \frac{1}{1 + e^{-(\alpha + \beta_1 \Delta Revenue + \beta_2 \text{ChurnRisk})}}$$

What We Learned

This project taught us that the future of BI is not "Chat-with-Data," but "Reasoning-over-Data." We learned how to leverage Salesforce Data 360 to create a "Single Source of Truth" that informs AI agents, preventing the "Hallucination" problem common in generic LLM implementations. By grounding Gemini in actual Tableau metadata, we achieved a $94\%$ accuracy rate in automated business recommendations. PulsePoint proves that when analytics and action collide, business moves at the speed of thought.