CasePilot

AI-Powered Decision Intelligence for Customer Operations

Transform unstructured business communications into explainable insights, intelligent prioritization, and actionable decisions using Google Gemini, Google Cloud, and NVIDIA RAPIDS.

---

Overview

Modern organizations generate an overwhelming amount of unstructured information—emails, requests, reports, alerts, approvals, and conversations.

While Generative AI can summarize this information, teams still face critical questions:

• What requires immediate attention?
• Which items carry the highest business risk?
• Where should resources be allocated first?
• Which decisions should be made today?

CasePilot is an AI-powered Decision Intelligence Platform that transforms unstructured business communications into structured intelligence, computes transparent business risk scores, prioritizes work based on explainable logic, and delivers actionable insights through an interactive dashboard.

Rather than replacing human decision-makers, CasePilot augments operational teams with AI-assisted prioritization and explainable recommendations.

Architecture

             Business Communications
                      │
                      ▼
              Data Ingestion Layer
                      │
                      ▼
          Google Gemini Structuring
                      │
                      ▼
          Structured Business Records
                      │
                      ▼
       Explainable Decision Intelligence
                      │
                      ▼
      Risk Classification & Prioritization
                      │
            ┌─────────┴─────────┐
            ▼                   ▼
     NVIDIA RAPIDS       Streamlit Dashboard
      GPU Analytics      Decision Console

---

Features

AI-powered Case Structuring

Google Gemini extracts structured intelligence including:

• Category
• Priority
• Sentiment
• Business Impact
• Root Cause
• AI Summary
• Recommended Action
• Confidence Score
• Named Entities

---

Explainable Risk Scoring

Instead of relying solely on LLM outputs, CasePilot combines AI with deterministic business rules to compute a transparent Health Score (0–100).

Factors include:

• Urgency
• Sentiment
• Business Impact
• Historical Context
• AI Confidence

Cases are then classified into:

• High Risk
• Medium Risk
• Low Risk

---

Interactive Decision Dashboard

The Streamlit dashboard provides:

• Executive KPI Cards
• Risk Distribution
• Issue Category Analytics
• Health Score Distribution
• AI-generated Operational Insights
• Searchable Case Explorer
• Detailed Case View
• Gemini-powered Resolution Suggestions
• AI-generated Response Drafts

---

NVIDIA RAPIDS Acceleration

To demonstrate enterprise scalability, CasePilot includes GPU-accelerated analytics using NVIDIA RAPIDS cuDF.

A synthetically expanded dataset (100K rows) is included to benchmark common analytical operations against pandas.

The benchmark performs:

• Aggregations
• GroupBy
• Sorting
• Category Analytics
• Risk Distribution

---

Technology Stack

Google Cloud

• Google Gemini
• Gmail API
• Cloud Storage (dataset hosting)

NVIDIA

• RAPIDS cuDF

Python

• Streamlit
• Plotly
• Pandas
• NumPy

---

Repository Structure

CasePilot/

├── app.py
├── benchmark.py
├── gmail_ingest.py
├── gemini_structure.py
├── scoring.py
├── export_results.py
├── upscale_data.py
├── generate_dummy_cases.py

├── CasePilot.ipynb

├── casepilot_scored_cases.csv
├── casepilot_structured_cases.csv
├── casepilot_scaled_100k.csv

├── requirements.txt
├── README.md
├── logo.png

└── casepilot-auth/
    └── authorize.py

---

Dataset

The repository contains three datasets:

File	Description
casepilot_structured_cases.csv	AI-structured customer support cases
casepilot_scored_cases.csv	Final explainable scoring output
casepilot_scaled_100k.csv	Synthetic dataset for RAPIDS benchmarking

The original benchmark was executed on a 2 million row synthetic dataset, which is intentionally excluded from the repository due to size constraints.

---

Running the Dashboard

Install dependencies

pip install -r requirements.txt

Launch the application

streamlit run app.py

---

Running the Benchmark

python benchmark.py

For GPU acceleration, install RAPIDS separately on a supported CUDA environment.

---

Explainable Health Score

Health Score is computed using a transparent rules engine.

Factors include:

• Urgency
• Business Impact
• Customer Sentiment
• Repeat Complaint Frequency
• AI Confidence

Unlike black-box AI scoring, every score can be traced back to explicit business logic.

---

Why CasePilot?

Traditional support dashboards answer:

What happened?

CasePilot answers:

What should we do next?

By combining LLM-based understanding with deterministic business rules, CasePilot helps support teams prioritize work, reduce escalations, and make faster operational decisions.

---

Future Improvements

• Vertex AI integration
• BigQuery analytics
• Multi-agent workflows
• Automatic SLA prediction
• Customer churn prediction
• Real-time Gmail monitoring
• Multi-channel support (Slack, Zendesk, Jira)

---

Built For

Google Cloud × NVIDIA AI Open Hackathon 2026

CasePilot demonstrates how Google Gemini and NVIDIA RAPIDS can be combined to build intelligent, scalable decision-support systems for customer operations.

---

License

MIT License.