How seifabroud/decision-agents-gemini works
Favorita is a sales decision-support platform: an offline pipeline prepares forecasts and analytical artifacts, which the Streamlit dashboard reads for exploration and reporting. Its agent workflow analyzes data, forecasts, recommends actions, checks cited numbers against replayable evidence, and produces reports; a separate assistant answers questions using tools and business-context retrieval. The README documents the pipeline and major dashboard capabilities; sampled source confirms key agent, tool, verification, and view relationships, but not every page’s wiring.
Main entry points: common.py Shared dashboard data · graph.py Analysis agent graph · tools.py Evidence tools · qa.py Question answering
Components 21
Data and forecasting
- Forecast pipeline
pipeline/build.py - Analysis artifacts
artifacts
Decision dashboard
- Dashboard entry
app/streamlit_app.py - Shared dashboard data
app/common.py - Overview
app/views/home.py - Store explorer
app/views/explorer.py - Forecast view
app/views/forecast.py - Agent results
app/views/agents.py - Data assistant view
app/views/assistant.py - Method view
app/views/method.py
Agent workflows
- Analysis agent graph
agents/graph.py - Evidence tools
agents/tools.py - Evidence verifier
agents/verify.py - LLM client
agents/llm.py - Question answering
agents/qa.py
Knowledge and reporting
- Business knowledge search
agents/rag.py - Business knowledge base
kb - Report generator
agents/report.py
Other components
- Decision maker
- Favorita CSV
- Gemini or Claude
Connections 26
- Decision maker → Dashboard entry: uses dashboard
README.md - Favorita CSV → Forecast pipeline: supplies sales
README.md - Forecast pipeline → Analysis artifacts: exports results
README.md - Dashboard entry → Analysis artifacts: reads artifacts
README.md - Overview → Shared dashboard data: loads shared data
app/views/home.py - Store explorer → Shared dashboard data: loads shared data
app/views/explorer.py - Forecast view → Shared dashboard data: loads shared data
app/views/forecast.py - Agent results → Shared dashboard data: sets up page
app/views/agents.py - Data assistant view → Shared dashboard data: sets up page
app/views/assistant.py - Method view → Shared dashboard data: sets up page
app/views/method.py - Shared dashboard data → Analysis artifacts: reads Parquet and JSON
app/common.py - Agent results → Analysis agent graph: runs analysis
app/views/agents.py - Agent results → Report generator: builds reports
app/views/agents.py - Analysis agent graph → Evidence tools: calls tools
agents/graph.py - Analysis agent graph → LLM client: requests generation
agents/graph.py - Analysis agent graph → Business knowledge search: retrieves context
agents/graph.py - Analysis agent graph → Evidence verifier: checks recommendations
agents/graph.py - Evidence tools → Analysis artifacts: reads analysis data
agents/tools.py - Question answering → Evidence tools: invokes selected tools
agents/qa.py - Question answering → LLM client: requests answers
agents/qa.py - Question answering → Business knowledge search: retrieves context
agents/qa.py - Question answering → Evidence verifier: checks answer numbers
agents/qa.py - Evidence verifier → Evidence tools: replays evidence
agents/verify.py - Business knowledge search → Business knowledge base: searches Markdown
agents/rag.py - Report generator → Evidence tools: reads analysis data
agents/report.py - LLM client → Gemini or Claude: calls provider
README.md