How kesavaai/vecturabi works
VecturaBI turns conversational questions into validated, typed query plans, executes them over in-memory Polars data, and presents results with visualizations, explanations, and audit details. The agent coordinates cache lookup, schema-aware parsing, validation, execution, and downstream analytics; the UI renders the returned result and supporting outputs. README capabilities are included where documented, but unsampled implementation details are represented only at their existing enclosing paths.
Main entry points: nodes.py Agent nodes · main.py BI dashboard · loader.py Data loader · chat.py Chat interface
Components 22
User experience
- BI dashboard
app/main.py - Chat interface
app/components/chat.py - Result views
app/components/data_table.py - Execution inspector
app/components/inspector.py
Agent workflow
- Agent graph
core/agent/graph.py - Agent nodes
core/agent/nodes.py - Query parser
core/agent/prompt_templates.py - Semantic cache
core/cache/semantic_cache.py
Query execution
- Data loader
core/data/loader.py - Sales dataset
data/enterprise_sales_data.parquet - Query AST
core/ast_engine/models.py - AST validator
core/ast_engine/validator.py - Polars compiler
core/ast_engine/compiler.py - SQL transparency
core/ast_engine/sql_translator.py
Analysis outputs
- Analytics coordinator
core/agent/nodes.py - Anomaly detection
core/analytics/anomalies.py - Diagnostics
core/analytics/diagnostic.py - Executive insights
core/analytics/insights.py - Next analysis
core/analytics/recommender.py - Briefing generator
core/analytics/briefing_generator.py - Chart engine
core/visualizer/chart_engine.py
Other components
- BI user
Connections 25
- BI user → Chat interface: asks question
- Chat interface → BI dashboard: submits query
- BI dashboard → Agent graph: invokes agent
- Agent graph → Agent nodes: dispatches steps
- Agent nodes → Data loader: reads schema
- Agent nodes → Semantic cache: looks up plan
- Agent nodes → Query parser: parses query
- Agent nodes → AST validator: validates AST
- Agent nodes → Polars compiler: executes query
- Agent nodes → SQL transparency: translates AST
- Data loader → Sales dataset: loads data
- Data loader → Agent nodes: provides schema
- Query parser → Query AST: builds AST
- AST validator → Query AST: sanitizes AST
- Agent nodes → Semantic cache: stores plan
- Polars compiler → Data loader: reads lazy data
- Agent nodes → Analytics coordinator: analyzes results
- Analytics coordinator → Anomaly detection: detects anomalies
- Analytics coordinator → Diagnostics: diagnoses issues
- Analytics coordinator → Executive insights: summarizes results
- Analytics coordinator → Next analysis: suggests analyses
- Analytics coordinator → Briefing generator: generates briefing
- BI dashboard → Chart engine: builds charts
- BI dashboard → Result views: renders results
- BI dashboard → Execution inspector: renders audit