Skip to main content

​
Quick Start Guide (Production Focus)

This guide is for enterprise teams standing up Simba Intelligence in a production (or production-like) environment. It intentionally avoids developer workstation / sample data instructions and instead walks you through the minimum sequence to become operational with your own infrastructure and data.

​
Overview of the Flow

  1. Deploy the platform
  2. Configure an approved LLM / AI provider
  3. Register at least one production data connector and create a data connection
  4. Generate an initial governed data source using the Data Source Agent
  5. Enable business users to explore via the Playground
  6. Validate operational readiness & plan next hardening steps

​
1. Deploy the Platform

Deployment steps (container orchestration, Helm, configuration files, secrets, scaling parameters, ingress, TLS, etc.) are documented in the dedicated Installation Guide. ➡️ Follow: Installation Guide When complete you should have:
  • Platform services healthy
  • Administrative (supervisor) credentials provisioned
  • Network access from the platform to required outbound AI endpoints and internal data sources

​
2. Configure an LLM / AI Provider

LLM configuration enables natural language understanding, semantic modeling, and embeddings. Use production credentials (not personal developer keys).
  1. Sign in with a supervisor / administrator role.
  2. Navigate to: /llm-configuration in the UI.
  3. Select your provider (e.g., Vertex AI, Azure OpenAI, Bedrock).
  4. Provide the required credentials (service account JSON, API key + endpoint, or similar) via the secure form.
  5. Enable the ncessary Chat + Embeddings services and save.
Configuration Reference (detailed parameters & environment keys): Use those sections for exact variable names, model identifiers, timeout and retry tuning options. Security & Ops Recommendations:
  • Store original secrets in your enterprise vault (e.g., HashiCorp Vault, AWS Secrets Manager); rotate per policy.
  • Use least-privilege IAM role for service accounts.
  • Monitor usage limits / quotas early to baseline cost.
(Optional) Multiple Providers: You can configure one primary provider at launch; add fallbacks later once baseline validation is complete.

​
3. Create a Data Connection

Connect an authoritative data source (e.g., Snowflake, PostgreSQL, SQL Server, BigQuery, etc.). Avoid pointing to non-governed sandbox datasets for your first production rollout.
  1. Navigate to: Data Connections in the UI.
  2. Click Create Connection.
  3. Choose connector type.
  4. Supply credentials (service user / managed identity preferred) + network parameters (host / account / warehouse / catalog / database / schema as applicable).
  5. (Recommended) Scope access to read-only analytical schemas for the initial rollout.
  6. Test Connection → Save.
  7. Allow initial metadata discovery to complete (time depends on catalog size).
Guidance by common platforms (generic):
  • Snowflake: Provide account locator / region, warehouse sized for concurrency testing, role with USAGE + SELECT.
  • PostgreSQL / SQL Server: Prefer read replica endpoints to avoid production OLTP impact.
  • BigQuery: Service account with dataset-level roles/bigquery.dataViewer and project metadataViewer.
More detail: Data Connection Management Guide

​
4. Create an Initial AI-Generated Data Source

Use the Data Source Agent to translate business intent into a governed data source definition over your connected warehouse.
  1. Navigate to: Data Source Agent (/data-source-agent).
  2. Select the connection you just registered.
  3. (Optional) Provide a clear objective (concise, business-oriented). Examples: 
    Build a revenue performance dataset with metrics: total revenue, revenue by region, revenue by product category, year-over-year growth, top 10 customers by trailing 90 day revenue.

    Create a customer health dataset including: active subscriptions, churn flags, MRR, expansion vs contraction amounts, tenure buckets.
  4. (Optional) List critical tables or schemas to prioritize (helps reduce exploration time).
  5. Start the agent. It will inspect schema metadata, propose transformations / joins, and assemble the data source definition.
  6. Review the generated fields; rename or hide any that should not be exposed. Approve to publish.
Operational Tips:
  • Keep the first data source narrow; optimize clarity over breadth.
  • Enforce naming conventions (e.g., revenue_total, customer_churn_rate).
  • Capture data lineage output for audit (export if needed).
More detail: Data Agent User Guide

​
5. Explore with the Playground

The Playground allows natural language querying and iterative exploration over approved data sources.
  1. Navigate to: Playground (/playground).
  2. Select the published data source created above.
  3. Ask domain-focused questions or use the suggested questions. Examples:
    • “What is total revenue this quarter vs the same quarter last year?”
    • “Show top 15 customers by trailing 90 day revenue and percent growth vs prior 90 days.”
    • “Break down subscription churn by region and customer segment.”
    • “List products with declining monthly revenue for the last 4 months.”
    • “Identify anomalies in daily order volume over the past 30 days.”
    • “Compare average customer lifetime value between enterprise and mid-market segments.”
Recommendation: Establish a review loop with data owners for early queries to validate accuracy. More detail: Playground User Guide

​
6. Validate Operational Readiness

Confirm the platform is stable, secure, and delivering expected value before scaling access. Readiness Checklist:
  • ✅ LLM provider status healthy; usage metrics monitored
  • ✅ At least one production data connection
  • ✅ Initial curated data source published and reviewed
  • ✅ Successful natural language queries returning accurate SQL and results
  • ✅ Access controls / roles applied (admin vs analyst vs viewer)
  • ✅ Logging & monitoring integrated (infrastructure + application + AI usage)
  • ✅ Backup / recovery strategy documented (see Operations Guide)
Next Hardening Steps:
  • Add additional connections (e.g., finance, customer success, product analytics) incrementally.
  • Implement tagging / classification for sensitive fields (PII, financial, etc.).
  • Configure performance monitoring & query cost dashboards.
  • Set up automated model / provider health alerts.

​
Troubleshooting (Production Context)

Common early issues: LLM Provider Authentication Fails
  • Verify service account / key has required model permissions.
  • Confirm outbound firewall rules allow provider endpoints.
Metadata Discovery Slow
  • Limit initial schema scope; exclude large archival schemas.
  • Schedule off-peak discovery windows for very large catalogs.
Unexpected Query Costs
  • Enable caching in each data source
  • Add row-level / column filters to high-volume tables.
Result Inconsistencies
  • Validate source table freshness; confirm no stale replicas.
  • Check semantic field definitions for unintended aggregations.
For deeper diagnostics use: Operations Guide, Performance Monitoring guide, and central logging stack.
You’re now ready to onboard additional teams. Keep the initial surface area small, validate quality, then scale connections, data sources, and user groups iteratively.