AI Data Analysis: A New Approach - Multi-engine Joint State Data Analysis
Organizations: Zhuhai National Taxation Bureau, China
Abstract
AI data analysis today is three fragmented worlds: LLMs understand semantics but cannot compute; BI aggregates precisely but is stateless; ML processes large-scale data on a stateful kernel but lacks semantics. This paper proposes a combination mechanism, named Multi-engine Joint State Data Analysis, that fills a still-open gap - the multi-engine joint context: the data context is embedded into LLM conversational analysis, where the LLM serves only as the semantic, reasoning, and orchestration layer (not computation) and connects four engines through the 1M context window - Python and R ML kernels, SQL over relational databases, Cypher over graph databases - yielding two multiplications, "semantics and precise computation" and "relational and graph", under enterprise-grade governance. The novelty is a four-dimensional coordinate system: horizontal (engine kinds), vertical (state flow - the watershed), spatial (business generalization), temporal (iterative convergence); a deterministic grounding method (one-shot injection of schema.json + notes.md, M+S) carries the spatial and temporal dimensions. Empirically, all four dimensions are substantiated: horizontally, four engines are quantified - SQL, Cypher (96.5% on CypherBench movie), ML in Python and R (TabReD weather); vertically, state flows across engines and grounding is reused across kernels; spatially, SQL generalizes across three business databases - AdventureWorks 91.3%, financial 95.3%, formula_1 94.6% (average), about 20-23 points above single-shot SOTA 74.45%; temporally, accuracy converges monotonically as notes accumulate. The core invariant - the separation of semantics from computation plus a cross-data-domain persistent context - does not expire as LLMs and agents evolve or engine forms change.
Figures & tables
| Config | Avg (3-round) | All-3 | Union ( 1 correct) |
|---|---|---|---|
| M+S M0 | 91.3% [86.8–95.2] | 85.7% [79.4–91.3] | 95.2% |
| M1 | 91.0% [86.8–94.7] | 83.3% [77.0–89.7] | 96.8% |
| M2 | 91.0% [86.8–94.7] | 82.5% [75.4–88.9] | 96.8% |
| M0 | 89.7% [84.9–93.9] | 82.5% [75.4–88.9] | 94.4% |
| S1 | 85.2% [79.4–90.5] | 80.2% [73.0–86.5] | 89.7% |
| S2 | 83.3% [77.5–88.6] | 75.4% [67.5–82.5] | 90.5% |
| Config | pass1 | pass2 | pass3 | Avg |
|---|---|---|---|---|
| M+S M0 | 90.5% | 92.1% | 91.3% | 91.3% |
| M1 | 91.3% | 88.9% | 92.9% | 91.0% |
| M2 | 91.3% | 91.3% | 90.5% | 91.0% |
| M0 | 92.1% | 88.9% | 88.1% | 89.7% |
| S1 | 86.5% | 84.9% | 84.1% | 85.2% |
| S2 | 81.7% | 82.5% | 85.7% | 83.3% |
| Dimension | single-shot best | pure-M best | M+S M0 |
|---|---|---|---|
| Avg accuracy | 85.2% (S1) | 91.0% (M1/M2) | 91.3% |
| All-3 | 80.2% (S1) | 83.3% (M1) | 85.7% |
| Cost (rounds / tokens) | 1 round / ~10–16k | 10.6 rounds / ~450k | 4.8 rounds / 275k |
| Drift-question ratio | 9.5% (S1) | 11.9 | 9.5% |
| Database | Convergence trajectory (zero ceiling) | Converged level |
|---|---|---|
| works_cycles | 69.8% (starting M0 round1 single-shot, old prompt/gold) 92.1% (round5 pure M) 91.3% (M+S avg) 97.6% gold-v4 | avg 91.3% / single-shot 97.6% (123/126) |
| financial | 75.5% 79.2% 84.9% (3-round convergence from zero) 84.9% (round 4) 95.3% grounding full (avg) 96.2% single-shot ceiling | avg 95.3% / single-shot 96.2% |
| formula_1 | 70.9% 73.8% 81.4% 83.1% (4-round monotonic) 93.6% (round5 single-shot) 94.6% grounding full (avg) 97.1% single-shot ceiling | avg 94.6% / single-shot 97.1% |
| Model | single-shot | M+S (avg / all-3) | S M+S gain |
|---|---|---|---|
| deepseek-v4-flash | 87.3% (S2) | 91.3% / 85.7% | +4pp |
| glm-5.2 | 77.8% (S2) | 91.8% / 85.7% | +14.0pp |
| glm-5.1 (GLM-5-Pro, 256K) | 48.4% / 54.0% (S1/S2) | 81.0% / 65.9% | +27~33pp |
| Round | notes / gold version | Full EX |
|---|---|---|
| Round 1 | v1 (start: initial schema + basic notes) | 85.8% |
| Round 3 | v3 (two rounds of failure root-cause rules: duplicate-name node dedup / grouping / NULL ordering) | 95.0% |
| Final | v4 (+ fixing 4 comparison-question gold cartesian products + correcting the Genre.name value rule) | 96.5% |
| Dimension | Engine / mechanism | Validated instance | Evidence strength |
|---|---|---|---|
| Horizontal | SQL RDB | works_cycles + financial + formula_1 three databases | quantitative (§4, strongest) |
| Horizontal | Cypher GraphDB | CypherBench movie 401 questions | quantitative (single graph, convergence trajectory + all-3) |
| Horizontal | ML Python kernel | Shanghai housing 270k rows + TabReD weather 16.95M | multiplication quantitative + cross-dataset reproduction |
| Horizontal | ML R kernel | rtorch MLP + float32 GBDT (TabReD weather) | capability-aligned with Python |
| Vertical | state flow | CSV shared layer + on-disk recovery | qualitative evidence |
| Vertical | grounding reuse | 370-table SQL R Python notes reuse | qualitative evidence |
| Dimension | Axis | Supporting evidence | Location |
|---|---|---|---|
| Horizontal | engine kinds | one instance per engine, four engines: SQL three databases / Cypher movie / ML Python / ML R | §5.1–5.3 |
| Vertical | state flow | CSV shared layer + notes reuse across kernels (370-table three kernels) | §5.4 |
| Spatial | business generalization | the same grounding holds across three databases (including two most unfavorable dirty databases) | §4 |
| Temporal | iterative convergence | convergence curve monotonically approaches, leading single-shot SOTA by ~20–23pp | §4.4 |
Data Analysis Agents
Data analysis agents use language models and tools to explore datasets, run analyses, and produce answers or reports grounded in evidence.
See all 117 papers