Legacy SQL to AI-Ready: A Pilot Path
You have a SQL Server database from 2012. It has 400 tables, inconsistent naming conventions, and documentation that was last updated when David Cameron was Prime Minister. Your CEO wants AI. Your board wants AI. Your competitors have AI. But your data is not ready. Or is it?
Week 1: Discovery
The KynticAI Discovery Agent connects to your database with read-only credentials once access has been approved. It scans table schemas, maps column types, identifies relationships, and generates a metadata profile. No data is copied. No ETL pipeline is built. Just a practical understanding of what you have.
The output is a Semantic Map: a visual representation of your data estate showing which tables contain customer context, which contain transactional history, and which contain operational metadata. Fields are automatically tagged with semantic hints — “this looks like an email address,” “this looks like a revenue figure,” “this looks like a timestamp.”
Week 2–3: Selector Configuration
Selectors are the core of KynticAI's mapping engine. They define how raw database fields translate into semantic attributes. The selector engine supports five mapping kinds:
DirectFieldMapping: A straight translation. The “email” column becomes the “contactEmail” attribute.
StringToEnumMapping: Free-text values mapped to controlled vocabularies. “Hot/Warm/Cold” becomes an engagementLevel score.
ThresholdClassification: Numeric values classified into bands. Revenue above £100k = “high value.”
WeightedScoring: Multiple fields combined into a composite score with configurable weights.
FormulaMetric: Custom calculations applied to raw data for derived metrics.
Week 4: Context Assembly
With selectors configured, the context layer begins generating Context Facts. Each fact includes the semantic attribute key, the resolved value, a confidence score (0.0 to 1.0), and a complete provenance chain tracing back to the source table, column, and record.
Context Snapshots assemble multiple facts into a point-in-time view of any entity. A snapshot for “Customer 12345” might include their churnRisk (0.34, confidence 0.91), conversionProbability (0.67, confidence 0.85), and preferredChannel (“email”, confidence 0.78) — all derived from your legacy SQL database without any data migration.
Week 5–6: Integration and Validation
Context Snapshots are exposed via GraphQL and REST APIs, plus TypeScript and C# SDKs. Your existing applications can query for context in a single API call. Your AI models receive structured, governed context instead of raw database dumps.
The validation phase uses the self-improving flywheel: as business outcomes occur (sales close, customers churn, support tickets resolve), the flywheel attributes results back to context signals and adjusts selector weights. The pilot goal is a governed context layer running on your legacy SQL database with clear evidence of which signals deserve production rollout.
What You Did Not Have to Do
You did not build a data warehouse. You did not hire an ETL engineering team. You did not migrate to a new database. You did not rewrite your applications. You did not send your data to a US cloud provider. You scoped KynticAI against what you already had, and the pilot created a concrete path to AI-ready context.