Forecasting, Causal Inference & AI-Assisted Analytics — a practitioner knowledge base for building and evaluating decision systems.
This repo documents frameworks, methods, and applied research across three areas of decision science. It is not a course log. It is a working reference built in public.
Covers the full methodology stack: statistical approaches (ARIMA family, ETS, Holt-Winters), machine learning and global models (gradient boosting with lag features, deep learning), and foundation models (zero-shot approaches such as Chronos). Includes backtesting design, accuracy metrics, prediction intervals, and decision thresholds.
Most analytics questions are causal, but most analytics methods are only descriptive. This pillar covers experimental design (A/B testing, RCTs), observational methods (difference-in-differences, regression discontinuity, synthetic controls), causal graphs and assumptions, and uplift/incrementality modeling.
Not building AI chatbots — evaluating whether AI-generated analysis is correct, useful, and decision-ready. Covers eval dataset design, answer rubrics, failure taxonomy (hallucination, wrong metric, overclaim), and observability basics.
intelligence-layer/
├── forecasting/
│ ├── statistical/ # ARIMA, SARIMA, SARIMAX, ETS, Holt-Winters, Prophet
│ ├── machine-learning/ # XGBoost/LightGBM with lag features, global models, deep learning
│ ├── foundation-models/ # Chronos, TimesFM, Moirai — zero-shot approaches
│ └── evaluation/ # Backtesting, accuracy metrics, prediction intervals
├── causal-inference/
│ ├── experimentation/ # A/B testing, RCTs
│ ├── observational/ # DiD, regression discontinuity, synthetic controls
│ ├── causal-graphs/ # DAGs, assumptions, do-calculus basics
│ └── uplift/ # Uplift modeling, incrementality
└── ai-evaluation/
├── eval-design/ # Benchmark questions, dataset construction
├── rubrics/ # Answer grading, correctness criteria
├── failure-taxonomy/ # Hallucination, wrong metric, overclaim
└── observability/ # Traces, guardrails basics
├── semantic-layer/ # Infrastructure literacy — metric governance, OSI, Omni
│ ├── frameworks/
│ └── omni/
└── growth-analytics/ # Application domain — PLG metrics, retention, revenue
└── frameworks/
└── resources/ # Annotated reading list and reference material
- Commits: Conventional Commits —
docs:,feat:,chore:,research:,fix: - Branches: topic-scoped —
docs/forecasting-evaluation,feat/causal-dag-framework,research/chronos-backtesting - Python: Poetry for dependency management
Built by Alejandro Berrizbeitia — Data Scientist, Time Series & Forecasting Professor at IE University, Senior Data Analyst at Kit.
