A secure, zero-data-movement solution for real-time customer sentiment analysis and summarization, fully deployed within the Snowflake Data Cloud.
This project integrates operational call transcript data with Generative AI to provide support managers with immediate, actionable insights into customer satisfaction and critical issues.
This project directly addresses the challenge of scaling call center quality assurance and customer risk management.
Manually reviewing high volumes of call transcripts is inefficient, leading to delayed responses to frustrated customers — ultimately causing customer churn and brand damage.
The primary goal is to transform unstructured call data into immediate, actionable intelligence:
- Risk Prioritization: Instantly identify negative sentiment calls to flag high-risk customers for urgent follow-up.
- Operational Efficiency: Generate concise AI summaries to quickly understand call context without reading the full transcript.
- Issue Routing: Automatically classify topics (e.g., Billing, Technical Support) to ensure fast and accurate escalation.
By running this analysis natively inside Snowflake, the solution guarantees:
- 🔒 Zero data egress — all processing occurs within Snowflake
- ⚡ High-speed inference using serverless AI
- 🧱 Enterprise-grade security and compliance
This project leverages a fully integrated, cloud-native stack for end-to-end AI-driven analytics.
| Component | Technology | Showcase Skill / Purpose |
|---|---|---|
| Application Hosting | Streamlit in Snowflake (SiS) | Interactive, secure data apps with zero data movement and native deployment inside Snowflake. |
| Data Engine | Snowflake Data Cloud | Centralized, secure storage for raw CALL_TRANSCRIPTS. |
| AI / NLP Engine | Snowflake Cortex AI | Serverless AI functions for sentiment, summarization, and topic classification. |
| Data Orchestration | Snowpark (Python) | Executes SQL + Cortex calls directly from Python for real-time analysis. |
The application leverages four on-demand AI functions powered by Snowflake Cortex — executed only when required to optimize cost.
| Feature | Cortex Function | Purpose | Cost Implication |
|---|---|---|---|
| Sentiment Analysis | SNOWFLAKE.CORTEX.SENTIMENT() |
Detects emotional tone (Positive, Negative, Neutral) | 💰 Low (input tokens only) |
| Summarization | SNOWFLAKE.CORTEX.SUMMARIZE() |
Generates 2–3 sentence summaries of long transcripts | 💰💰 Moderate–High |
| Classification | SNOWFLAKE.CORTEX.CLASSIFY_TEXT() |
Categorizes calls into topics (e.g., “Billing”, “Technical Support”) | 💰 Low–Moderate |
These features empower managers to focus on what matters — customer emotion, context, and resolution priority.
This project demonstrates strong engineering and data integration skills within the Snowflake ecosystem:
- Native Cloud AI Integration: Combined multiple serverless AI functions (
SUMMARIZE(),SENTIMENT(),CLASSIFY_TEXT()) inside the Streamlit workflow — no external models or APIs needed. - Secure Parameterized SQL: Implemented variable passing between Python and Snowflake using the
$$delimiter, preventing SQL injection and invalid identifiers. - Dynamic Classification Logic: Constructed flexible SQL arrays (
ARRAY_CONSTRUCT) to enable adaptive topic tagging. - Cost-Efficient Design: AI functions run only when triggered by the user, aligning with token-based pricing for Cortex.
Ensure your Snowflake account has:
- ✅ Cortex AI functions enabled
- ✅ Role with
USAGEprivileges on Cortex - ✅ A virtual warehouse (e.g.,
X-Small) with auto-suspend for cost control
Run the following SQL scripts inside Snowsight or SnowSQL:
-- Create database and schema
CREATE OR REPLACE DATABASE STREAMLIT_DB;
CREATE OR REPLACE SCHEMA STREAMLIT_SCHEMA;
-- Create table for transcripts
CREATE OR REPLACE TABLE CALL_TRANSCRIPTS (
CALL_ID STRING,
AGENT_NAME STRING,
CUSTOMER_NAME STRING,
CALL_DATE DATE,
TRANSCRIPT TEXT
);
-- Load sample transcript data
INSERT INTO CALL_TRANSCRIPTS VALUES
('C001', 'Amit', 'John Doe', '2025-10-01', 'Customer reported poor service response...'),
('C002', 'Maya', 'Jane Smith', '2025-10-02', 'Customer appreciated the agent support...'),
('C003', 'Raj', 'Michael Lee', '2025-10-03', 'Billing issue unresolved, customer upset...');
The repository is organized for quick deployment within the Streamlit in Snowflake environment:
/Call_Center_Cortex_Streamlit
├── README.md
├── streamlit_app.py # Core Streamlit app logic
└── Data_Setup/
├── create_tables.sql # SQL for table creation
└── sample_data_load.sql # Sample transcript inserts
Example: 