SonyProfessionalIn Development

Sales Data Agent - Enterprise AI with Snowflake Intelligence

Built an enterprise-grade AI agent (SALES_AI) using Snowflake Cortex Intelligence that enables business users, analysts, and new joiners to query sales data in plain English without SQL knowledge. The agent combines Cortex Analyst for text-to-SQL conversion, Cortex Search for RAG-based semantic search over customer support tickets, and custom tools for automated email notifications. Implementing AI Observability framework using TruLens to ensure reliable, grounded, and trustworthy responses for business-critical decisions. The agent orchestrates multiple data sources, generates verified SQL queries, creates visual analytics, and provides natural language insights — transforming traditional BI into conversational, agentic analytics.

LinkedIn Post

Problem Statement

Business users traditionally depend on data analysts or technical teams to write complex SQL queries for insights, creating bottlenecks and delays in decision-making. Additionally, LLMs can hallucinate, and in business decisions, accuracy is non-negotiable. The challenge was to enable non-technical users to query sales data using natural language while ensuring every AI-generated response is reliable, grounded, and trustworthy for business-critical decisions.

My Approach

Developed SALES_AI, an enterprise AI agent using Snowflake Cortex Intelligence that orchestrates multiple tools: (1) Cortex Analyst for converting natural language questions to verified SQL using YAML-based semantic models that map business terminology to database schemas, (2) Cortex Search for hybrid vector + keyword search over unstructured customer support tickets (RAG), and (3) custom stored procedures for automated actions like email notifications. Implemented multi-step agentic reasoning that plans, iterates, and combines insights from structured and unstructured data sources. Deployed AI Observability framework using TruLens to evaluate answer relevance, correctness, groundedness, cost, and latency — ensuring enterprise-grade reliability with systematic batch evaluations across 40+ test questions.

Key Outcomes

Enabled business users to query sales data in plain English without SQL knowledge
Automated complex multi-step analytics combining structured (SQL) and unstructured (RAG) data
Generated verified SQL queries with natural language explanations and visual charts in seconds
Identified sales trends and patterns (e.g., Fitness Wear sales surge from $2.2M to $6.8M in July, then drop to $1.0M in August)
Explained internal terminology and company-specific abbreviations automatically
Implemented AI Observability framework using TruLens for trustworthy, grounded responses
Built semantic models bridging business language to technical database schemas
Enforced enterprise governance with RBAC, data masking, and row-level security
Reduced analyst workload by automating routine sales analysis and reporting
Currently designing scalable evaluation framework for continuous AI agent improvement

Tech Stack

Snowflake Cortex IntelligenceSnowflake Cortex Analyst (Text-to-SQL)Snowflake Cortex Search (RAG)TruLens (AI Observability)Snowflake Arctic Embed (Embeddings)Meta Llama (LLM)YAML (Semantic Models)SQLPythonSnowflake RBACStored Procedures