SAIDA

SAIDA is a Python framework for deterministic data analysis.

Its core contract is simple:

• input: AnalysisPlan
• output: AnalysisResult

I built SAIDA to make BI dashboarding and day-to-day analysis easier to automate, easier to test, and easier to plug into apps.

What You Get

• canonical AnalysisPlan execution
• deterministic validation before compute
• standardized AnalysisResult output
• pluggable source, compute, and output adapters
• optional prompt and LLM helpers that sit on top of the core

The main runtime flow is:

• Dataset -> AnalysisPlan -> Validate -> Execute -> AnalysisResult

Quick Start

1. Load a dataset

from saida.sources import CSVSource

dataset = CSVSource(
    "examples/datasets/support_tickets_500.csv",
    context_path="examples/contexts/support_tickets_500.md",
).load()

2. Execute an authored plan

from saida import Saida
from saida.core.contracts import AnalysisPlan, PlanStep

engine = Saida()

plan = AnalysisPlan(
    task_type="descriptive",
    rationale="Count tickets by team.",
    expected_result_name="group_row_counts",
    expected_result_shape="table",
    steps=[
        PlanStep(
            step_id="count_rows_by_group",
            tool_family="duckdb",
            action="count_rows_by_group",
            method_id="count_rows_by_group",
            family="aggregation_grouping",
            parameters={"group_by": ["team"]},
            description="Count tickets by team.",
        )
    ],
)

result = engine.execute_plan(dataset, plan)

print(result.response["result"]["name"])
print(result.response["result"]["value"])

3. Render the result

json_payload = engine.render_output(result, output_format="json")
summary_text = engine.render_output(result, output_format="summary")

Optional Prompt Frontend

Prompt handling is supported, but it is optional.

If you want prompt-to-plan generation:

from saida import PromptAnalysisFrontend

frontend = PromptAnalysisFrontend()
plan = frontend.plan(dataset, "How many tickets do we have by team?")
result = frontend.analyze(dataset, "How many tickets do we have by team?")

Prompt and LLM features are not the compute layer. They are convenience tooling that generates candidate plans or optional summaries.

Built-In Interfaces

Sources

• CSV
• Excel
• JSON
• pandas
• SQL sources such as SQLite, PostgreSQL, and MySQL

Compute

• DuckDB
• metadata/profile-backed execution
• statsmodels-backed execution

Output

• JSON
• summary text

Current Scope

SAIDA is strongest today as a framework for:

• aggregation and grouping
• tabular retrieval
• ranking
• verification checks
• schema and metadata inspection
• distinct value analysis
• time-based analysis
• period comparison
• statistical inference
• diagnostic workflows

Reserved ML-facing APIs still exist, but forecasting and model training are not the main stable feature area yet.

Read Next

• Docs Index
• Architecture
• API Usage
• Schema Spec
• Prompt Playbook
• Prompt Family Catalog