Implementation
Client onboarding
A five-step implementation process that integrates your relational databases with the natural-language query interface. Automated schema extraction and vectorization do the bulk of the work, and most steps require no input from your team.
Five steps, from access to go-live.
Connection and credentials setup
Connection strings, credentials and IP whitelisting.
- Validate network connectivity to each database instance
- Initialise the connection pool manager
- Register the database instances into the system registry
Connection parameters: host, port, user, password and database name
Read-only credentials configured for each database
Whitelisting of our service IP addresses
Schema ingestion and semantic vectorization
Automated schema extraction and vector embedding.
- Extract relational metadata via INFORMATION_SCHEMA and system catalogs, covering tables, columns, types, keys and relationships
- Synthesise raw schema definitions into enriched semantic JSON using schema interpreters
- Embed the semantic schema into isolated vector indices for query classification and routing
No client input is required. This step is fully automated.
Prompt customisation and routing calibration
Domain terminology, KPIs and multi-database routing.
- Tailor database agent prompts to domain-specific business terminology, KPIs and multi-table join logic
- Calibrate similarity thresholds so incoming queries route accurately to the target database
Optional: five to ten frequently asked business questions or reporting metrics, to accelerate prompt tuning
Query validation and internal QA
Accuracy, sanitisation and execution safety.
- Execute internal test suites to verify SQL generation accuracy, syntax sanitisation and execution safety
- Calibrate the response generator to produce clear textual and tabular summaries
Optional: confirmation of the expected answers for any sample questions you provided
Client review, UAT and go-live
Joint walkthrough, then production handover.
- Provision user access to the conversational interface and REST API endpoints
- Conduct a live walkthrough with client stakeholders using real operational queries
- Incorporate final adjustments and transition the solution to production support
Stakeholder availability for the walkthrough, and confirmation of performance acceptance
Deliverables and responsibilities.
Client input is limited to database access and, optionally, a set of sample queries. All remaining items are delivered by us.
| Deliverable | Owner | Description and requirements |
|---|---|---|
| Database credentials and access | Client | Read-only access credentials and network IP whitelisting for each target database. |
| Sample queries | Client | Optional. Five to ten frequent business questions or key reporting metrics, provided to accelerate prompt tuning. |
| Schema vector store | DataConverse | Automated schema extraction, semantic enrichment and isolated vector indexing per database. |
| Tuned SQL generation pipeline | DataConverse | Domain-optimised system prompts, verified routing thresholds and SQL sanitisation. |
| Production interface and API | DataConverse | Full deployment of the conversational interface and backend API endpoints, with SLA support. |
Frequently asked questions.
How long does it take?
It depends on how many databases are in scope and how complex their schemas are. A single straightforward database can be completed in a day. We give you a firm schedule after the first call, once we know what we are connecting to.
Which databases are supported?
PostgreSQL on port 5432, MySQL on port 3306 and Microsoft SQL Server on port 1433. Document stores, knowledge graphs and NoSQL sources are on the roadmap and are not part of a standard implementation.
Is a semantic layer required before we begin?
No. The semantic layer is generated at step two. Raw schema metadata is synthesised into enriched semantic JSON and embedded automatically, so no data modelling work is required from your team.
How much client time does the implementation require?
Credentials and IP whitelisting at step one, an optional set of sample questions, and stakeholder availability for the walkthrough at step five. Steps two, three and four need little or nothing from you.
What happens when the generated SQL is incorrect?
The generated query is displayed alongside every answer, so an error is visible rather than hidden. Steps three and four exist to identify and correct these before go-live, using internal test suites executed against your own questions.
Is an API provided, or only the conversational interface?
Both. Step five provisions access to the conversational web interface and to REST API endpoints, allowing the engine to be integrated with existing internal tools.
What level of database access is required?
Read-only access on every database. The engine does not issue write operations, and a read-only role enforces this constraint at the database level.
Begin the onboarding process.
Implementation begins once database access and IP whitelisting are in place. Step one can start as soon as they are.