One Search. Every Answer.
Databases hold the figures. Documents hold the reasons. JKN Tech built an enterprise retrieval system that answers from both — exact numbers computed in your database, context cited from your own documents.
Your Question
What was Q3 revenue growth for Product X, and what strategic factors caused the deviation from our projected budget?▊
Text-to-SQL · Exact Figures
Q3 revenue growth computed directly from the warehouse schema — no LLM arithmetic.
RAG · Cited Context
Deviation drivers retrieved from earnings transcripts and board decks — with source citations.
The Problem
Your Data Lives in Two Silos
Modern enterprise data sits in two distinct worlds, each demanding a different analytical paradigm. Standard tools can only reach one of them.
Structured Data
Relational databases and data warehouses. Transaction ledgers, inventory counts, time-series metrics.
Exact numeric precision, schema enforcement, filtering and aggregation.
Blind to qualitative context and unindexed narrative text.
Reached by SQL queries.
Unstructured Data
PDF reports, emails, policy manuals, support tickets, call transcripts.
Rich semantics, nuance, and the qualitative rationale behind decisions.
Poor at exact calculations, aggregations, and real-time figures.
Reached by vector search and RAG.
The Insight Gap
"What was our Q3 revenue growth for Product X, and what strategic factors caused the deviation from our projected budget?"
Standard systems fail this question. A pure RAG pipeline struggles with exact financial calculation. A pure SQL pipeline returns the totals but cannot explain the qualitative why. The answer lives on both sides — and so does the retrieval.
The Solution
Hybrid AI Retrieval Architecture
A dual-engine framework that merges the deterministic precision of Text-to-SQL with the semantic flexibility of RAG — orchestrated by a Context & Intent Router.
database Text-to-SQL Pipeline
Converts natural language into validated SQL against your structured databases.
- check Schema linking & dynamic table selection
- check Dialect translation: PostgreSQL, Snowflake, BigQuery
- check Sandboxed syntax & execution validation
- check Read-only execution, column-level permission checks
manage_search RAG Pipeline
Extracts narrative, contextual knowledge from your unstructured documents.
- check Multi-stage chunking & metadata tagging
- check Hybrid vector search: dense semantic + sparse BM25
- check Cross-encoder reranking for relevancy scoring
- check Contextual compression to trim token overhead
route Context & Intent Router
Classifies every query and picks the optimal execution path — cutting latency and token cost instead of running everything through both pipelines.
- calculate Quantitative intent → Text-to-SQL
- notes Qualitative intent → RAG
- sync_alt Composite intent → parallel hybrid execution
Built for the People Asking the Questions
No query language. No hunting between BI dashboards and document folders. Executives, operations, sales, and compliance staff ask in plain language — and get figures with their reasons attached.
Zero Math Hallucination
Numbers execute in the database engine via SQL. LLMs never compute totals.
Dual Attribution on Every Answer
The underlying SQL query and direct document citations travel with the response.
Reduced Search Latency
Simple queries skip unnecessary processing. The router sends each request down the shortest path.
Why It Works
Hybrid Beats Either Engine Alone
Combining Text-to-SQL and RAG resolves the core limitation of using either technique in isolation.
| Evaluation Dimension | Pure Text-to-SQL | Pure RAG | JKN Tech Hybrid System |
|---|---|---|---|
| Numerical Accuracy | Deterministic — accurate to database state | Prone to LLM calculation hallucinations | Deterministic SQL accuracy for all metrics |
| Qualitative Depth | None — cannot read free text | High — retrieves narrative context | High — pairs numbers with narrative context |
| Query Flexibility | Limited to predefined schemas | Flexible across any written text | Comprehensive across all corporate assets |
| Data Recency | Real-time query execution | Limited by indexing schedule | Real-time database reads + indexed docs |
| Trust & Verification | Transparent SQL query inspection | Direct citation of source documents | Dual attribution: SQL query + doc citations |
Use Cases
Where the Hybrid Pays for Itself
The same architecture serves every role that has ever needed a number and the story behind it.
Executive & Financial Reporting
"What was our EBITDA margin expansion in Q2, and what management commentary was provided regarding rising logistics costs?"
SQL pulls exact quarterly figures from the warehouse; RAG retrieves the reasoning from earnings-call transcripts and board decks.
Supply Chain & Operations
"How many units of Part #4821 are in stock across European warehouses, and what are the force majeure terms in our vendor contract with Supplier Y?"
Real-time inventory counts from the ERP, clause specifics retrieved from the vendor contract PDF.
Customer Success & Contract Auditing
"What is Customer Z's current annual recurring revenue, and what are their custom SLA response commitments?"
Subscription details from the CRM, non-standard terms extracted from scanned agreement documents.
Regulatory & Compliance Auditing
"Which audit findings from last quarter remain unfulfilled, and what total financial penalties do they carry under current risk framework rules?"
Penalty thresholds from financial risk models, mapped to qualitative audit report summaries.
Enterprise Value
What the Hybrid System Delivers
Zero Math Hallucination
Every numerical query executes directly in your database engine. LLMs never compute totals, so calculation errors cannot enter your reporting.
Reduced Search Latency
The intent router keeps simple database or document queries off unnecessary processing stages. Fast paths stay fast.
Auditable Decisions
Every answer carries its evidence: the SQL query that produced the figures and footnoted citations for every source document.
Unified Knowledge Access
One natural-language interface over BI and knowledge-management systems. The friction between the teams disappears.
Your Data Already Holds the Answers.
Put one interface in front of it. JKN Tech designs, deploys, and manages the hybrid retrieval stack on infrastructure you control — your data never leaves your perimeter.