COURSE · BLD · I014

Data Analysis with Local AI

Learn data analysis with local ai through RunLocalAI's practical lens: data analysis, pandas, visualization and nlp, hardware fit, runtime settings, verification habits and local-vs-cloud tradeoffs.

18 chapters10hBuilder trackBy Fredoline Eruo
PREREQUISITES
  • B002
  • B003

Course I014: Data Analysis with Local AI

Why this course exists

Data analysis workflows involve repetitive decisions: which visualization fits this data type, what statistical test applies to this distribution, how to handle missing values correctly. These decisions consume time and require statistical knowledge that many practitioners lack deeply. Local AI changes this equation by providing contextual guidance throughout the analysis process without sending data to external servers.

Traditional data analysis tools automate computation but leave decision-making to humans. A spreadsheet calculates a mean instantly but offers no guidance on whether the mean is the appropriate measure. A visualization library renders any chart but does not suggest which chart reveals the pattern in the data. Local AI bridges this gap by acting as an analysis companion that understands both the data and the analytical techniques available.

This course teaches practical patterns for integrating local AI into data workflows using Ollama and Python libraries. The focus is on real workflows with real failure modes: models that hallucinate statistical terms, SQL generation that produces syntactically valid but semantically wrong queries, and visualization suggestions that ignore data types. Students learn to recognize these failure modes and correct them.

What you will know after

  • Configure Ollama for data analysis assistance with appropriate model selection
  • Use AI guidance to plan exploratory data analysis strategies for unfamiliar datasets
  • Generate profile reports that identify data quality issues automatically
  • Convert natural language questions into SQL queries using local models
  • Build interfaces that let non-technical users query data through conversation
  • Automate chart selection based on data types and analytical goals
  • Apply appropriate statistical tests using AI-generated guidance
  • Design and validate hypothesis tests with local model assistance
  • Debug common AI-assisted analysis failures before they reach stakeholders
CHAPTERS
  1. 01AI for Data AnalysisLocal AI augments analysis by providing technique guidance without sending data externally, but requires verification since models can suggest inappropriate methods for specific data situations.20 min
  2. 02AI-Guided EDAAI-guided EDA provides structure and recommendations, but data characteristics must be verified before applying each suggested technique.20 min
  3. 03Automated Data ProfilingAutomated profiling identifies data issues, but AI interpretation connects findings to analytical implications and remediation strategies.20 min
  4. 04Text-to-SQLText-to-SQL reduces the technical barrier to data access, but generated queries require validation against actual schema and execution results.25 min
  5. 05Natural Language QueriesNatural language query interfaces enable non-technical access to data, but must handle ambiguity explicitly and maintain conversation context across multiple turns.20 min
  6. 06Automated VisualizationAutomated visualization selects appropriate chart types based on data characteristics and analytical goals, but generated code requires validation for correctness and effectiveness.25 min
  7. 07Chart RecommendationsChart recommendation systems suggest not just individual visualizations but complete strategies including sequences, encodings, and aesthetic choices.20 min
  8. 08Statistical AnalysisAI guidance helps select appropriate statistical techniques based on data characteristics and provides interpretations that make results accessible to non-statistical audiences.20 min
  9. 09Hypothesis TestingHypothesis testing requires appropriate test selection, correct implementation, and assumption verification. AI guidance helps navigate each step while understanding limitations.25 min
  10. 10Correlation AnalysisSpearman correlation and Cramér's V handle non-normal distributions and categorical data where Pearson fails. Always lag-test correlations before claiming causation.20 min
  11. 11Time Series AnalysisResampling and rolling windows transform raw time series into meaningful patterns. ACF/PACF plots guide model selection for forecasting tasks.20 min
  12. 12Trend DetectionCombine multiple methods—MA smoothing for visualization, linear regression for quantification, CUSUM for change detection—to build reliable trend analysis.20 min
  13. 13Anomaly DetectionCombine statistical methods (Z-score, IQR) with ML approaches (Isolation Forest) and LLM-based explanation for complete anomaly analysis pipeline.20 min
  14. 14Natural Language InsightsStructured prompting with specific output format requirements enables reliable extraction of entities, sentiment, and answers from unstructured text without fine-tuning.20 min
  15. 15Data StorytellingStructure precedes visualization. Build the narrative arc before selecting charts—each visualization should serve a specific point in the story.20 min
  16. 16Report GenerationSeparate report generation from analysis logic. Store raw results first, then render them into multiple formats (Markdown, HTML, PDF) without re-running analysis.20 min
  17. 17Multi-Source AnalysisDefine a canonical schema before loading data. Map all sources to this schema during load rather than patching mismatches later.20 min
  18. 18Data Analysis Platform ProjectProduction platforms separate concerns—ingestion, analysis, and reporting are independent modules that communicate through well-defined interfaces and data structures.30 min
← All coursesStart chapter 1 →