Public-facing record of Samyama-related research publications. Manuscript sources, reviewer feedback, and submission timelines live in the private samyama-research repository.
All preprints under the author listing , newest last. Dates are first submission; v is the latest revision on arXiv as of 2026-10-07. The # column carries the internal paper number where one exists, because the benchmark table below references papers by that number — a dash means the preprint has no internal number, not that it is unnumbered work.
# arXiv Date v Cat Title
1 2603.08036 2026-03-09 v3 cs.DBSamyama: A Unified Graph-Vector Database with In-Database Optimization, Agentic Enrichment, and Hardware Acceleration 4 2603.15080 2026-03-16 v3 cs.DBOpen Biomedical Knowledge Graphs at Scale: Construction, Federation, and AI Agent Access with Samyama Graph Database 8 2605.12204 2026-05-12 v2 cs.DBGraph-Grounded Optimization: Rao-Family Metaheuristics, Classical OR, and SLM-Driven Formulation over Knowledge Graphs 3 2605.26874 2026-05-26 v3 cs.DBKnowledge Graphs as the Missing Data Layer for LLM-Based Industrial Asset Operations — 2606.15600 2026-06-14 v1 cs.DBWhen Does q-error Predict Plan Regret? Three Regimes of Cardinality-Estimation Error — 2606.16341 2026-06-15 v1 cs.LGFiltered ANN as a Phase Transition: When Selectivity-Estimation Error Causes Plan Regret — 2606.18138 2026-06-16 v1 cs.DBThe Value of Adaptivity in LSM Bloom-Filter Tuning: A Log-Law and a Two-Clock Frontier — 2606.18187 2026-06-16 v1 cs.DBGroup Commit Self-Clocks: Why Tuning Is Unnecessary Above a Device-Set Load Threshold — 2606.20539 2026-06-18 v2 cs.DBCaching for Dollars, Not Hits: An Exact Offline Reference for Cloud-Egress Caching and the Crossover That Decides When It Pays — 2606.22423 2026-06-21 v1 cs.DBWhen Is a Columnar Scan Bandwidth-Bound? A Decode-Throughput Law and Its Cross-Hardware Validation — 2606.22419 2026-06-21 v3 cs.CLKnowledge-Graph Grounding Helps LLMs Only for Out-of-Training Knowledge: A Controlled Study on Clinical Question Answering — 2606.24677 2026-06-23 v1 cs.DBOne Index for Subsumption and Roll-up across Time, Geography, and Ontology — 2607.00728 2026-07-01 v2 cs.DBWhen to Repair a Graph ANN Index: A Matched-Budget Negative Result, and the Interpolated-Baseline Trap That Hid It — 2607.01182 2026-07-01 v2 cs.DBThe Decode-Work Law: Margin-Governed, Provably-Exact Spatial Joins over Compressed Geometry — 2609.00378 2026-07-17 v1 cs.CLRemovable and Irreducible: A Token-Cost Ledger for the Multilingual Tokenization Tax — 2609.00381 2026-07-18 v1 cs.DBBounded, Indeterminate, or a Bug: A Condition-Aware Oracle for Differential Testing of SQL Aggregates — 2609.23032 2026-09-19 v1 cs.DBSame Pattern, Different Answer: A Reference Semantics and Divergence Map for GQL and SQL/PGQ Path Patterns — 2609.29894 2026-09-24 v2 math.PRA Computer-Assisted Proof of Speed Monotonicity for the Biased Random Walk on a Galton-Watson Tree Beyond the Known Range
Fourteen of the eighteen are cs.DB. Two are cs.CL, one cs.LG, and one — 2609.29894 , a computer-assisted proof about biased random walks on Galton-Watson trees — is math.PR and unrelated to Samyama. It is listed because this page tracks the author’s arXiv output, but it is not part of the database work and nothing in this book depends on it.
# Title Venue Status
5a Federated Biomedical KGs (Demo) GRADES-NDA @ SIGMOD 2026 Submitted 5b Federated Biomedical KGs (In-Use) ISWC 2026 In-Use Track Submitted 6 MCP Tools vs Text-to-Cypher aiDM @ SIGMOD 2026 Submitted 7 Billion-Edge KG Federation on Commodity Hardware arxiv preprint → VLDB 2027 In preparation — VLDB 2027 Systems Paper VLDB 2027 Industry Track In preparation (extends Paper 1)
# Venue target Status
1 arxiv preprint → VLDB 2027 Preprint at v3; v3 incorporates v1.0.0 + hero-run results (187.5M nodes / 1.29B edges / 99% pass, 2026-04-30) 3 NeurIPS 2026 E&D Preprint at v3 4 arxiv preprint → VLDB / ICDE 2027 Preprint at v3 (Drug Interactions KG, BiomedQA 98%) 8 arxiv preprint Preprint at v2
Benchmark Papers Results Hardware
LDBC SNB Interactive (21/21) 1, VLDB All pass, SF1 Mac Mini M4 LDBC Graphalytics (28/28) 1, VLDB All pass, XS–S Mac Mini M4 LDBC FinBench (40/40) 1, VLDB All pass Mac Mini M4 AssetOpsBench (139 scenarios) 3, 6 MCP 99% · T2C 83% · GPT-4 65% Mac Mini M4 BiomedQA (40 questions) 4, 5, 6 MCP 98% · T2C 85% · GPT-4o 75% AWS g4dn.4xlarge Biomedical 100-query (74M nodes, 1B edges) 7, VLDB 96 / 100 pass AWS r6a.8xlarge Public Health 40-query (305K nodes) 7 40 / 40 pass MacBook Pro Cricket 100-query (37K nodes) 1 87 / 100 Mac Mini M4 Mega Benchmark (138M nodes, 1.22B edges, 11 KGs) 5b, 7, VLDB 500 / 500 pass AWS r7i.16xlarge Hero Run (187.5M nodes, 1.29B edges, 1000 queries) 7, VLDB 99% pass (~$1.85)AWS r7i-class spot (2026-04-30) Paper 8 demo suite (7 problems, 19.7K–7.78M nodes) 8 BMWR 4/7, Rao-1 2/7 wins; OR-tools 2–3× on linear/MILP Laptop / single-node