Q

xAQUA

2.0

For Government
Sign In
Get Demo
⌨️ ConverseSQL · Enterprise SQL Generation Engine Semantic-Grounded Embedded across xAQUA

Natural language in. Production SQL out.

The SQL reasoning engine at the core of the xAQUA platform — built on the semantic layer, governed by design.

ConverseSQL turns plain-English questions into production-ready SQL across diverse target data sources. It is grounded by the xAQUA semantic layer, not generic LLM guessing — so the SQL it produces uses tables that exist, joins your team has approved, and dialects your engines understand. Proven in production on enterprise schemas with 500+ tables and 50+ columns per table on average.

500+ tables in production 50+ avg columns/table 12+ SQL dialects 4 embedded modules
Demo Against Your Schema → See the Architecture
📄 Download the ConverseSQL brochure PDF · 4 pages
⌨️
ConverseSQL
Semantic resolution → governed SQL → target dialect
Live
💬 "active members enrolled in the COLA-eligible service tier, by employer plan, last 5 fiscal years"
target → Snowflake | resolved across 47 tables
1
2
3
4
5
6
7
8
9
10
11
-- Active COLA-eligible members by employer plan · FY trend SELECT ep.plan_code, ep.plan_name, fy.fiscal_year, COUNT(DISTINCT m.member_id) AS active_members FROM dim_member m JOIN dim_service_tier st ON st.tier_key = m.tier_key JOIN dim_employer_plan ep ON ep.plan_key = m.plan_key JOIN dim_fiscal_year fy ON fy.fy_key = m.enrollment_fy_key WHERE st.cola_eligible = TRUE AND m.status = 'ACTIVE' GROUP BY 1,2,3 ORDER BY 3 DESC, 4 DESC;
✓ semantic-resolved cola_eligible · active_member governance applied · dialect: snowflake · est. 4.3M rows scanned
ConverseSQL · production scale
500+
Tables / Schema
50+
Avg Columns / Table
12+
Target SQL Dialects
4
Embedded xAQUA Modules
VPC
Private Deployment
Why Enterprise SQL Generation Is Hard

Generic AI doesn't know your schema.

Real enterprise data does not live in textbook 3NF. It lives in 500+ table warehouses, wide fact tables with 50+ columns, decades of legacy naming, multiple SQL dialects, and dozens of governance rules that are nowhere in the DDL. A general-purpose copilot generates SQL that looks right and silently returns the wrong number.

ConverseSQL is built differently. Every prompt resolves first against the xAQUA semantic layer — the catalog of business entities, governed metrics, join graph, security policies, and dialect mappings that defines what your data actually means. The SQL it emits is the SQL your team would have written, scaled across the breadth of your enterprise estate.

Resolved before generated. Governed before executed. Production from day one.

The Real Problem

Four reasons enterprise text-to-SQL usually fails.

01 · SCALE
Too many tables to fit in a prompt
A 500-table schema doesn't fit in any context window. Naive RAG retrieval misses the joins. ConverseSQL resolves the right slice via the semantic graph, every time.
02 · WIDTH
Wide tables with cryptic columns
50+ columns per table, half of them flag_27 or amt_lcl_v2. Without business meaning attached, the model picks the wrong column. The semantic layer makes meaning explicit.
03 · DIALECT
12 dialects, one prompt
Snowflake date math, BigQuery arrays, Postgres JSON, Oracle's outer-join syntax. Generated SQL has to be valid in the target — not a generic dialect that needs rewriting.
04 · GOVERNANCE
Rules not in the DDL
"Active member" is a 3-condition filter. "Net revenue" excludes specific channels. These are policy, not schema. ConverseSQL inherits them from the semantic layer, automatically.
How ConverseSQL Works

From a sentence to a query you can ship.

Five stages, one engine. The semantic layer does the heavy lifting before any token of SQL is generated.

01 · INPUT Natural Language User · API · Agent "top members by COLA…" 02 · SEMANTIC RESOLUTION xAQUA Semantic Layer Entities · 500+ tables resolved Metrics · governed definitions Join graph · approved paths Policy · row / column / mask Glossary · business synonyms 03 · GENERATION SQL Engine + Dialect Adapter CTE planning · join inference window / date / JSON syntax style profile · formatter 04 · GOVERNANCE Safety Gate RBAC · audit log DML confirmation PII masking applied 05 · EXECUTE Target Engines Snowflake · Databricks BigQuery · Redshift DuckDB · MotherDuck Postgres · SQL Server Oracle · Synapse · Trino + legacy RDBMS EMBEDDED IN → ConverseDataIQ Athyna Composer xAQUA Reeve + Composer Migration · API · CLI
Semantic Layer · ground truth
Generation · governance
I/O · execution
Embedded surfaces
One Interface · Many Targets

Diverse target data sources. One semantic interface.

Enterprise data does not standardize on one engine. Finance lives in Snowflake. The lakehouse runs on Databricks. There is still a Postgres from 2018 powering the operational app, and an Oracle warehouse no one is allowed to touch. ConverseSQL writes for all of them.

Switch targets with a parameter. The semantic resolution stays the same — only the dialect, functions, and execution path change. The same prompt that returned a result against the lakehouse yesterday returns the same result against the warehouse today.

  • Native generation for cloud warehouses, lakehouses, RDBMS, and DuckDB
  • Dialect-correct date, window, JSON, array, and geospatial functions
  • Cross-dialect translation for migration projects
  • Same semantic layer, same governance, different executor
❄️
Snowflake
CLOUD DW
⚡
Databricks
LAKEHOUSE
🟦
BigQuery
CLOUD DW
🟥
Redshift
CLOUD DW
🦆
DuckDB
ADL · NATIVE
🌊
MotherDuck
SERVERLESS
🐘
Postgres
RDBMS
🟧
Oracle
RDBMS · LEGACY
🔷
SQL Server
RDBMS · SYNAPSE
Use Cases · Full Data Lifecycle

Anywhere a workflow can be expressed in SQL.

ConverseSQL is the SQL generation engine the rest of xAQUA depends on. The same engine powers analytics, harvesting, prep, ETL, migration, and validation.

📊
Analytics & Ad-hoc Queries
Analysts · BI · finance
Describe a metric in English. Get a governed query in your dialect, joined correctly across the semantic graph, ready to ship to a dashboard or report.
🪣
Data Harvesting
Source extraction · profiling
Generate extraction queries against legacy and operational sources. Reverse-engineer table contents, sample data, and produce profiling SQL without writing it by hand.
🔄
Data Migration & Modernization
Cross-dialect translation
Translate Oracle PL/SQL to Snowflake. Rewrite Redshift to BigQuery. Move legacy procedures to modern lakehouse SQL. Flag the queries that need human review.
🧪
Migration Testing & Validation
Reconciliation · row-count parity
Generate parity, reconciliation, and parallel-run queries between source and target. ConverseSQL writes both sides so the validation is symmetric, not hand-coded twice.
🧹
No-Code Data Preparation
Cleansing · enrichment
Powering Athyna's no-code studio. Users describe a transformation in plain English; ConverseSQL generates the underlying SQL — pushdown to the warehouse, full lineage tracked.
⚙️
Transformation & ETL Logic
Pipeline development · dbt-style
Composer uses ConverseSQL to draft transformation logic for ETL pipelines. Engineers describe the model; the engine produces dialect-correct, idempotent transformation SQL.
⌨️ Core Platform Capability

Built into the xAQUA platform.

ConverseSQL is not a side feature. It is the SQL generation engine embedded across xAQUA — wherever a user, agent, or pipeline needs governed SQL from intent.

💬
ConverseDataIQ
Chat with Data
The conversational analytics surface for business users. Every "chat with data" question routes through ConverseSQL to produce governed, semantically-grounded answers from enterprise sources — not generic LLM guesses.
ConverseSQL role: natural-language → governed analytics SQL
🧰
Athyna
Cloud Data Studio · Prep & Transformation
The no-code / low-code data preparation and transformation studio. Users describe a cleansing rule, a join, a filter, an enrichment in plain English — Athyna calls ConverseSQL to generate the underlying SQL with full pushdown to the target warehouse.
ConverseSQL role: no-code prep → executable transformation SQL
⚙️
Composer
ETL · Pipelines · Orchestration
The pipeline development environment. ConverseSQL accelerates ETL build, transformation logic creation, migration translation, and pipeline test generation. Engineers describe; the engine drafts; humans review, version, and ship.
ConverseSQL role: pipeline intent → idempotent transformation & test SQL
📚
xAQUA Reeve
Data Product Catalog · Governed Consumption
The catalog of certified data products for enterprise consumption. Reeve uses ConverseSQL for data product exploration, governed access, and self-serve consumption — letting consumers query certified products in plain English without breaking governance.
ConverseSQL role: product browse → policy-aware consumption SQL
One generation engine. Four embedded surfaces. The same semantic resolution, governance, and dialect handling whether the user is an analyst chatting with data, a business user preparing a dataset, an engineer building a pipeline, or a consumer browsing a data product. That is what makes ConverseSQL a platform capability — not a tool.
Production-Scale Control

Governance, security, and deployment built in.

ConverseSQL was built for regulated, high-scale environments — pension funds, government agencies, financial services, healthcare. Governance is not a setting; it is the architecture.

🔐
Private VPC Deployment
Runs entirely inside the customer boundary. Schema, prompts, and generated SQL never leave. No data trains any external model.
  • AWS · Azure · GCP · on-prem
  • Air-gap and disconnected modes
  • Bring-your-own LLM (private GPT-OSS, Llama)
👤
RBAC & Row-Level Security
Generated SQL respects the user's permissions. ConverseSQL will not write a query the user is not allowed to run, and applies row / column / mask policy on the way out.
  • SSO / SAML / OIDC integration
  • Role-aware semantic resolution
  • PII masking honored automatically
📜
Full Audit & Lineage
Every generation, explanation, and execution logged: prompt, resolved entities, generated SQL, executor, user, result row count. Full lineage for compliance and review.
  • Per-user, per-prompt audit trail
  • Lineage of resolved tables / columns
  • Export to SIEM & governance tools
⛔
Safe Execution Gates
Destructive DML requires explicit confirmation. Long-running queries surface plan and cost estimates first. Dry-run mode validates before any execution.
  • DDL / DML confirmation gates
  • Cost & row-count preview
  • Dry-run + EXPLAIN integration
🧠
Semantic Layer Native
No separate metadata. No second source of truth. ConverseSQL resolves directly against the xAQUA semantic layer — the same one Athyna, Composer, Reeve, and ConverseDataIQ use.
  • Single shared semantic model
  • Live updates · no rebuild
  • Glossary, synonyms, business rules
🔌
Embed via API
Use ConverseSQL inside xAQUA modules, in your own applications, or in agent workflows. REST and gRPC endpoints, Python and TypeScript SDKs, and an LLM Gateway envelope.
  • REST · gRPC · Python · TS SDKs
  • LLM Gateway integration
  • Agent & workflow embedding

Bring your schema. See your SQL.

Watch ConverseSQL generate production SQL against your semantic layer, your dialect, your governance rules. Thirty minutes, live against your environment — including the wide tables and the legacy joins.

Request a ConverseSQL Demo → Explore the Semantic Layer 📄 Download Brochure (PDF)