CRO Analyst v3 is a production-grade web application designed to centralize, analyze, and make accessible A/B testing experiment data for conversion rate optimization teams. The platform transforms scattered experiment data into actionable insights through intelligent data visualization, natural language querying, and graph-based relationship analysis.

This application serves as a source for experiment tracking, enabling teams to quickly discover patterns, learn from past experiments, and make data-driven decisions about future tests.

The Challenge

Before CRO Analyst v3, experiment data lived in disconnected spreadsheets, making it difficult to:

Track experiment outcomes across different teams, verticals, and geographies
Identify winning patterns that could be replicated across similar contexts
Answer questions quickly without writing complex SQL queries
Understand relationships between experiments, change types, and business metrics
Maintain data freshness as new experiments concluded

The solution needed to be:

Accessible to non-technical stakeholders (PMs, analysts)
Transparent in its data sources and query logic
Scalable to handle growing experiment volumes
Automated to reduce manual data entry overhead

Solution Architecture

Core Features

1. Automated Data Ingestion

Google Sheets Integration: Automatically syncs experiment data from live Google Sheets
Scheduled Imports: Daily cron jobs ensure data stays current without manual intervention
Manual Import Options: Support for CSV uploads and on-demand sheet imports
Data Normalization: Robust parsing and validation of dates, numbers, and text fields
Screenshot Management: Automatic retrieval and storage of experiment screenshots from Google Drive

2. Intelligent Experiment Discovery

Stats Dashboard: Monthly overview with key metrics, top winners, and breakdowns by vertical/geo
Experiment Detail Pages: Comprehensive views showing hypothesis, lessons learned, revenue impact, and visual evidence
Graph Visualization: 3D interactive Neo4j graph showing relationships between experiments, change types, elements, and business dimensions
Similarity Matching: AI-powered discovery of related experiments based on shared characteristics and outcomes

3. Natural Language Query Interface

Ask AI Feature: Query experiment data using plain English
Dual-Mode Querying: Automatically routes questions to SQL (Postgres) for data queries or Cypher (Neo4j) for relationship queries
Transparent Results: Shows generated queries and source data for full transparency
Safe Query Execution: Built-in sanitization prevents SQL injection and enforces read-only access

4. Graph-Based Analytics

Neo4j Integration: Mirrors experiment data into a graph database for relationship analysis
Pattern Discovery: Identifies connections between change types, UI elements, verticals, geographies, and outcomes
Visual Exploration: Interactive 3D force-directed graphs powered by Three.js
Similar Experiment Recommendations: Uses graph relationships to suggest related experiments

Technical Stack

Frontend:

Next.js 15 (App Router) with React 19
TypeScript for type safety
TailwindCSS for styling
Three.js + react-force-graph-3d for 3D visualizations

Backend & Data:

PostgreSQL with Prisma ORM (canonical data store)
Neo4j Graph Database (relationship analysis)
NextAuth.js for Google OAuth authentication
OpenAI API for embeddings and LLM-powered query generation

Infrastructure:

Vercel for hosting and serverless functions
Vercel Cron for scheduled data imports
Google Sheets API for data source
Google Drive API for screenshot management

Technical Highlights

1. Hybrid Database Architecture

The application uses a dual-database approach:

PostgreSQL: Stores the canonical experiment data with full schema, supporting complex queries and aggregations
Neo4j: Maintains a graph mirror optimized for relationship traversal and pattern discovery

This architecture allows the system to leverage the strengths of both relational and graph databases, providing fast analytical queries while enabling rich relationship exploration.

2. AI-Powered Query Generation

The "Ask AI" feature uses LLM-based query generation with multiple safety layers:

Query Sanitization: Validates SQL/Cypher queries before execution
Read-Only Enforcement: Blocks any write operations
Automatic Mode Selection: Intelligently chooses between SQL and graph queries based on question intent
Result Summarization: Provides natural language summaries of query results
Transparency: Always shows the generated query and source data

3. Automated Data Pipeline

The import system is designed for reliability and scalability:

Idempotent Imports: Safe to run multiple times without duplicating data
Batch Processing: Handles large datasets in manageable chunks
Error Recovery: Gracefully handles partial failures
Progress Tracking: Job state persisted in database for resumability

4. Graph Similarity Algorithm

The system uses a sophisticated algorithm to find similar experiments:

Primary: Uses explicit SIMILAR_TO relationships when available
Fallback: Calculates similarity based on overlap of:
- Change types
- UI elements modified
- Vertical/geography/brand
- Target metrics
Ranking: Orders by overlap score and monthly revenue extrapolation
Limiting: Returns top 6 most similar experiments

User Experience

Stats Dashboard

Automatically selects the latest completed month with data
Displays key metrics in card format: experiments run, concluded, launched
Shows top winners by revenue impact
Provides breakdowns by vertical and geography
Features an interactive 3D graph visualization
Paginated experiment table with filtering

Experiment Detail Pages

Complete experiment metadata
Hypothesis and lessons learned
Revenue impact calculations
Screenshot gallery
Links to external tools (e.g., Optimizely)
Graph of similar experiments
Context-aware navigation (remembers origin page)

Ask AI Interface

Simple text input for natural language questions
Mode selection (Auto, SQL-only, Graph-only)
Real-time query execution
Expandable sections showing:
- Generated SQL/Cypher queries
- Source data tables
- Visual graph representations (for graph queries)
Clickable experiment links for easy navigation

Do You Need Help Building Intelligent Data Analytics Solutions?

If you're working with large datasets, complex relationships, or need to make data more accessible to your team, I can help design and build custom analytics platforms that combine the power of SQL and graph databases. Let's discuss how to transform your data into actionable insights.

Book a Free Strategy Call

Security & Operations

Authentication

Google OAuth via NextAuth.js
Optional email allowlist for access control
Session-based authentication for manual imports

API Security

Internal API key for scheduled imports
Service account tokens for Google API access
Query sanitization prevents injection attacks
Read-only query enforcement

Monitoring

Health check endpoint (/api/health)
Database connectivity monitoring
Error logging and tracking
Vercel serverless function monitoring

Impact & Results

For Teams

Time Savings: Reduced time to find experiment information from hours to seconds
Pattern Recognition: Graph visualization reveals winning patterns across verticals and geographies
Knowledge Sharing: Centralized repository of hypotheses and lessons learned
Data-Driven Decisions: Easy access to historical performance data

For Analysts

Query Transparency: See exactly what queries are being run
No SQL Required: Natural language interface makes data accessible
Relationship Discovery: Graph analysis reveals unexpected connections
Reproducible Insights: All queries and results are traceable

For the Organization

Centralized Source: Eliminates confusion from multiple data sources
Automated Updates: Daily syncs ensure data freshness
Scalable Architecture: Handles growing experiment volumes
Future-Proof: Extensible design supports new features and integrations

Key Learnings

Dual Database Strategy: Combining relational and graph databases provides both analytical power and relationship insights
Transparency Builds Trust: Showing generated queries increases user confidence in AI-powered features
Automation Reduces Friction: Scheduled imports eliminate manual data entry overhead
Visualization Aids Discovery: 3D graphs reveal patterns that tables cannot
Safety First: Multiple layers of query sanitization are essential for AI-generated code

Future Enhancements

Potential areas for expansion:

Advanced filtering and faceted search
Experiment recommendation engine
Automated insight generation
Integration with additional experiment platforms
Team collaboration features
Custom dashboard creation

Conclusion

CRO Analyst v3 demonstrates how modern web technologies, intelligent data architecture, and user-centered design can transform complex data into actionable insights. By combining automated data ingestion, graph-based analytics, and natural language querying, the platform makes experiment data accessible to everyone while maintaining the rigor and transparency that analysts require.

The project showcases expertise in:

Full-stack Next.js development
Database architecture (relational + graph)
AI/LLM integration with safety considerations
Data pipeline automation
Interactive data visualization
Production deployment and operations

Technologies: Next.js, React, TypeScript, PostgreSQL, Neo4j, Prisma, NextAuth, OpenAI API, Google APIs, Vercel, Three.js, TailwindCSS

Role: Full-stack Developer & Architect

Timeline: End 2025 / Start 2026

CRO Analyst v3 — Intelligent Experiment Discovery Platform