A professional-grade data analysis agent built as a ChatGPT Custom GPT. Designed for automated EDA, hybrid AutoML modeling, PII detection, leakage detection, and governance — with dynamic adaptation to Beginner, Expert, and Executive users.
- Dynamic user expertise adaptation (Beginner / Expert / Executive)
- Two-mode automated EDA (Light and Full)
- Mandatory preprocessing pipeline before modeling
- Hybrid AutoML with self-correction fallback loop
- PII and sensitive field auto-detection
- Data leakage detection and categorization
- Uncertainty and confidence reporting on every model
- Multi-file handling (join, append, or separate analysis)
- Session context management
- Text intelligence via TF-IDF or embeddings
- Hardened confidentiality and security protection
- ChatGPT Custom GPT (GPT-4)
- Python (Pandas, Scikit-learn, TF-IDF)
- Logistic Regression, RandomForest, Ridge, Linear Regression
- One-hot, frequency, and target encoding strategies
The agent is powered by a versioned, compact instruction set. See the full specification:
📄 instructions/agent-instructions-v7.md
See changelog.md for full version history from v6C through v7.
Self-Correction Loop — automatically detects model failure and attempts a single structured fallback before stopping and explaining why.
Executive Output Structure — leads with conclusion, follows with evidence, ends with action. Never methodology first.
Leakage Detection — scans for post-event indicators and categorizes features as Safe, Suspicious, or High-risk before any modeling begins.
Multi-File Handling — detects shared keys, matching schemas, or structural differences and routes accordingly with row count validation.
Paul Orlando Creative Technologist | AI Agent Developer | Data Analytics 🌐 paulforlando.com