Principal Data Scientist | Causal Inference & Marketing Science | AI/LLM Systems | Data Science Leadership
I lead a team of 8 Data Scientists that add business value through advanced modeling. My team and I build advanced statistical models, machine learning systems, and AI-powered tools that help organizations make data-driven decisions. My work focuses on causal inference, marketing mix modeling, LLM agentic systems, and deep learning for time series analysis.
SAGE - Strategic AI-Guided Explorer for Marketing Performance. An AI-powered copilot for Marketing Mix Modeling with natural language interface.
- 🤖 Agentic AI System with OpenAI function calling
- 💬 Natural language interface - ask questions in plain English
- 📊 Automatic visualizations - generates Plotly charts on demand
- 🎯 Budget optimization - SLSQP-based allocation across channels
- 🧠 RAG-powered insights - ChromaDB knowledge base with semantic search
- ⚡ Real-time analysis - instant answers to MMM questions
https://SAGEinsights.streamlit.app | 🚀 Live Demo | 📖 GitHub
llm-copilot - Production-ready agentic system for Marketing Mix Modeling with LLM orchestration, RAG, and code execution.
- 🎭 Agentic System: OpenAI function calling for tool orchestration
- 📚 RAG Implementation: ChromaDB + OpenAI embeddings for semantic search
- 🔧 Dynamic Code Execution: Safe Python sandbox for on-the-fly analysis
- 📈 Response Curve Fitting: Automatic Hill curve generation with deepcausalmmm
- 🗃️ Knowledge Base: Stores curves, benchmarks, glossary, and best practices
- 🔌 Multi-Database Support: 10+ connectors (Databricks, Snowflake, BigQuery, etc.)
pip install git+https://github.com/adityapt/llm-copilot.git Tech: OpenAI GPT-4, ChromaDB, LangChain patterns, Pandas, Plotly
Creator and maintainer of DeepCausalMMM - an open-source Python package for Marketing Mix Modeling with causal inference and deep learning.
- 🧠 GRU-based temporal modeling for automatic adstock and lag learning
- 🔗 DAG causal discovery to reveal channel interdependencies
- 📉 Response curves with Hill saturation for budget optimization
- 🌍 Multi-region modeling with learnable region-specific effects
- ⚡ Performance: 93% holdout R², 3.6% train-test gap
pip install deepcausalmmm📖 Documentation | 🎯 Quick Start | 📝 Examples
- Agentic system design with function calling
- Retrieval-Augmented Generation (RAG)
- Prompt engineering & Chain-of-Thought
- Vector databases (ChromaDB, Pinecone)
- LLM orchestration patterns
- Marketing Mix Modeling (MMM)
- Directed Acyclic Graphs (DAGs)
- Causal structure learning
- Treatment effect estimation
- Instrumental variables
- Time series forecasting
- Recurrent Neural Networks (GRU, LSTM)
- PyTorch model development
- Bayesian inference
- Ensemble methods
- Brand Science
- Marketing Mix Modeling
- Churn prediction
- A/B testing & experimentation
- Budget optimization
- Executive Insights
-
SAGE AI Copilot (2025) - AI-powered MMM assistant with agentic system
- Deployed on Streamlit Cloud
- Natural language interface for marketing analytics
- RAG-powered insights with ChromaDB
-
LLM-Copilot (2025) - Production agentic system for MMM
- OpenAI function calling orchestration
- Dynamic code execution sandbox
- Multi-database connectivity
-
DeepCausalMMM (2025) - Advanced MMM with causal inference
- Published on PyPI with 1.0.17+ versions
- Comprehensive documentation on ReadTheDocs
- 28 comprehensive tests with 100% pass rate
- JOSS paper submitted
- ORCID: 0009-0008-9495-3932
- Focus areas: Causal inference, marketing analytics, LLM systems, time series modeling, Marketing mix modeling
- ✅ Built production agentic system with OpenAI function calling
- ✅ Implemented RAG with ChromaDB for semantic knowledge retrieval
- ✅ Created natural language interface for MMM analysis
- ✅ Integrated automatic budget optimization algorithms
- ✅ Deployed live Streamlit application
- ✅ Added 10+ database connectors (Databricks, Snowflake, etc.)
-
✅ Added non-linear response curves with Hill saturation
-
✅ Implemented proportional allocation for accurate scaling
-
✅ Enhanced Hill parameter constraints (slope ≥ 2.0)
-
✅ Integrated 14+ interactive visualizations
-
✅ Achieved 93% holdout R² with 3.6% performance gap
- Installation:
pip install deepcausalmmm - Documentation: deepcausalmmm.readthedocs.io
- License: MIT (fully open source)
- Python: 3.9+ compatible
- Framework: PyTorch 2.0+
- Installation:
Currently exploring:
- LLM agent architectures and multi-agent systems
- RAG optimization techniques and hybrid search
- Transformer architectures for time series
- Causal discovery algorithms (NOTEARS, PC, GES)
- Bayesian deep learning for uncertainty quantification
- Multi-task learning for marketing applications
- Federated learning for privacy-preserving MMM
"Build AI/ML systems that are both theoretically rigorous and practically useful. Combine machine learning with causal inference to create tools that explain why, not just what."
I believe in:
- 🤖 AI for Good: Building LLM systems that augment human decision-making
- 📖 Open Science: Making research accessible and reproducible
- 📚 Documentation: Clear guides that help others learn and contribute
- 🤝 Community: Collaborative development and knowledge sharing
- 🎯 Impact: Solving real problems with elegant solutions
- 🤖 LLM agentic systems and RAG applications
- 🔬 Causal inference research and applications
- 📊 Marketing analytics and MMM projects
- 🧠 Deep learning for time series
- 📦 Open source data science tools
- 📚 Technical writing and documentation
Feel free to reach out if you're working on something interesting in these areas!
