The 9-Layer Semantic Engine

Most NL-to-SQL tools send your question straight to an LLM. Echo builds a deep understanding of your schema first – then writes SQL that actually works.

The Pipeline

Nine layers of semantic intelligence, applied before a single token of SQL is generated.

1
2
3
4
5
6
7
8
9

Schema Linking

Maps 1,056+ column names to human-readable business terms. BELNR_CONCAT becomes “Document Number”, NETWR becomes “Net Value”.

Entity Resolution

Resolves business entities to database values. “Hyderabad” maps to plant code “1200”, “Q3” maps to fiscal period range.

Synonym Expansion

Understands that “PO” means “Purchase Order”, “GRN” means “Goods Receipt Note”, “AP” means “Accounts Payable”.

Type Awareness

Detects mismatched types. When amounts are stored as TEXT, Echo applies CAST operations automatically.

Relationship Context

Builds a TableRelationshipGraph and uses BFS to find optimal JOIN paths across 500+ table schemas.

Business Metric Templates

82 pre-defined metrics with SQL templates. DSO, revenue, margin, aging — computed correctly every time.

PowerBI DAX Integration

Reverse-engineers 274 DAX measures from your existing PowerBI reports to ensure dashboard consistency.

Computed Patterns

Encodes fiscal year logic, regional tax calculations, currency conversions, and business-specific formulas.

Few-Shot Selection

Uses RAG over a Gold Dataset of verified queries to find the most relevant examples for each new question.

25-Loop Self-Correction

Echo doesn’t just generate SQL — it verifies it. Every query goes through up to 25 iterative refinement loops.

Generate

LLM produces SQL using the full semantic context from all 9 layers.

Execute

Query runs against your database in read-only mode. Results are validated.

Validate & Correct

If the query fails, Echo diagnoses the error and regenerates with corrections.

Circuit Breaker

After repeated failures, the loop stops. No infinite retries or runaway costs.

Why 95%+ accuracy matters

Tool Accuracy Schema Intelligence Self-Correction Zero Data Movement ERP Native
AX Echo 95%+ 1,056+ fields 25 loops
Generic NL2SQL ~65% None

Ready to see 95%+ accuracy on
your schema?