Research
Papers & patents
The technical foundations behind Kaveon's deterministic natural-language analytics, compiled context, curation, and adaptive routing.
Research documents explain methods, experiments, and product direction. Consult
STATUS.md and feature guides before treating a described capability as shipped.Technical papers
| Paper | Focus | Repository source |
|---|---|---|
| Data Language Model | Self-compiling semantic context and deterministic question resolution | Read paper |
| DLM Curation at Scale | Precomputed answer tiers, HLL cuboids, incremental refresh, and serving | Read paper |
| Adaptive Context Routing | Validity scoring and selection between context and live data | Read paper |
| Deterministic NL→SQL | Template patterns, fuzzy matching, dataset selection, and chart choice | Read paper |
| Kaveon vs Trino | Distributed SQL architecture, maturity, differentiation, and benchmark proof requirements | Read comparison |
| Kaveon Engine vs Fabric SQL endpoint | Direct-lake execution compared with the managed Lakehouse SQL serving boundary | Read comparison |
| Kaveon vs HTAP platforms | Snowflake Unistore, Databricks LTAP, TiDB X, and Fabric compared on engines, storage authority, licensing, and maturity | Read comparison |
| KaveonDB scorecard | Per-dimension ratings against Trino for analytics and PostgreSQL for the record store, from measured evidence | Read scorecard |
Patent disclosure
The adaptive-context-routing disclosure records the invention, claims, embodiments, and prior-art delta. It is a legal/technical disclosure, not an API contract or deployment guide.
Reading paths
- Product practitioner: DLM guide → Freshness Algorithm → DLM Curation at Scale.
- Data engineer: Deterministic NL→SQL → Adaptive Context Routing → Architecture.
- Researcher: DLM paper → Curation paper → routing paper → patent disclosure.
The Markdown sources remain the canonical long-form editions and render directly on the repository host. Studio documentation provides the operational interpretation and route into each topic.