Karthik RameshContact
All workOpen-source reference build

Governed Analytics Copilot

A governed analytics copilot for natural-language SQL, forecasting, approvals and evidence-backed business decisions.

For
Analysts, operators and decision-makers working with governed business data.
Ownership
Product framing, Governance architecture, Backend and frontend delivery, Evaluation and security
System
FastAPI, Next.js, DuckDB, SQLGlot, PostgreSQL, MLflow
Analytics copilot workspace showing dataset analysis, generated SQL and an approval gate
End-to-end analytics workspace using the repository's synthetic demonstration data.Synthetic demonstration data
The workflow problem

What had to change

Business teams need faster answers from data, but generated SQL should never receive unrestricted authority over sensitive datasets.

The product response

How the product behaves

The copilot turns questions into reviewable SQL, visual analysis, forecasts and executive reports while preserving ownership, policy and execution budgets.

User journey

From first signal to inspectable outcome

  1. 01Select an owned dataset
  2. 02Ask a business question
  3. 03Review generated SQL
  4. 04Approve the exact query
  5. 05Inspect results and forecast
  6. 06Export evidence
Architecture

The control path

The model participates inside a product boundary. Typed software, policy and human review retain authority where consequences matter.

  1. Workspace

    Dataset ownership and business question

  2. Generation

    Natural language to constrained SQL proposal

  3. Governance

    AST allowlist, RBAC and immutable approval binding

  4. Execution

    DuckDB budgets for rows, bytes, time and memory

  5. Decision

    Visualisations, forecasts and executive reporting

Product decisions

Choices that shape the system

01

SQL is untrusted input

AST validation and physical-table allowlists precede execution.

02

Approval binds exact intent

SQL and dataset version are fingerprinted before execution.

03

Materialisation is budgeted

Rows, bytes, time and memory have explicit limits.

Evaluation and reliability

What keeps the product inspectable

Controls

  • Owner-scoped datasets
  • Immutable approval fingerprints
  • Execution budgets
  • Audit trail

Verified evidence

  • Natural-language analytics
  • Forecasting workflow
  • Governed SQL execution

Current limitations

  • The included runtime is a local-first reference build, not a production data warehouse.
  • Demonstration forecasts use synthetic datasets and are not business performance claims.
Governed analytics copilot SQL approval gate with exact query review and deterministic local-provider label
Exact-query approval surface using synthetic customer-churn data.Synthetic demonstration data
Analytics copilot revenue forecasting workspace showing a three-month forecast over synthetic monthly revenue
Actual forecasting surface populated with a synthetic three-month revenue fixture.Synthetic demonstration data
Karthik’s ownership

Product framing · Governance architecture · Backend and frontend delivery · Evaluation and security

GitHub repository View all work