Papers, code, and methods from the FAOS Research Programme — applied work on neurosymbolic enterprise AI, agent verification, ontology design, and context engineering. Built in public. Shipped where it matters.
Peer-grade work, open code, and honest status. Posted papers link to arXiv; the rest show their current status and reproducibility code while they finish review.
RA-1arXiv
Dynamic Coordination Strategy Selection
Thanh Luong Tuan
When should enterprise multi-agent systems use consensus, debate, synthesis, or a single agent? Across a frozen 1,440-output matrix spanning six industries and four model families, the answer is to route coordination strategy by problem class as a calibrated default — near-best routing holds in every model arm, even where a strict exact-winner rule does not.
@article{luong2026coordination,
title = {Dynamic Coordination Strategy Selection for Enterprise Multi-Agent Systems},
author = {Luong Tuan, Thanh},
year = {2026},
journal = {arXiv preprint},
eprint = {2606.00804},
archivePrefix = {arXiv},
note = {FAOS Research Programme, RA-1}
}
RA-3arXiv
Neurosymbolic Enterprise AI
Thanh Luong Tuan, Abhijit Sanyal
A grounded architecture for enterprise agents that fuse symbolic ontologies with neural reasoning — closing the precision gap between LLM-only systems and rule-based pipelines without sacrificing flexibility.
@article{luong2026neurosymbolic,
title = {Neurosymbolic Enterprise AI},
author = {Luong Tuan, Thanh and Sanyal, Abhijit},
year = {2026},
journal = {arXiv preprint},
eprint = {2604.00555},
archivePrefix = {arXiv},
note = {FAOS Research Programme, RA-3}
}
RA-4arXiv-ready
Empirical Bounds of Ontological Context
Thanh Luong Tuan
A pre-registered cross-model study of ontological context in enterprise agents. The universal context-efficiency theory fails; what survives is more useful: model-domain calibration, insurance-specific length sensitivity, framing audits, and context-budget discipline instead of context maximalism.
@article{luong2026empiricalbounds,
title = {Empirical Bounds of Ontological Context: A Pre-Registered Cross-Model Study of Length, Framing, and Domain Effects in Enterprise LLM Agents},
author = {Luong Tuan, Thanh},
year = {2026},
journal = {Preprint pending arXiv submission},
note = {FAOS Research Programme, RA-4. arXiv v1.0 source package prepared; public repository artifacts post at canonical arXiv release.}
}
RA-6In review
Pre-Deployment Assurance for Enterprise AI Agents
Thanh Luong Tuan, Abhijit Sanyal
A pre-deployment verification framework for enterprise agents — an operational envelope the agent is certified to operate within, test scenarios derived automatically from the industry ontology, and a machine-verifiable trust certificate that binds the agent version to its evidence.
@article{luong2026verification,
title = {Toward Pre-Deployment Assurance for Enterprise AI Agents: Ontology-Grounded Simulation and Trust Certification},
author = {Luong Tuan, Thanh and Sanyal, Abhijit},
year = {2026},
journal = {Submitted to MDPI Computers (in review)},
note = {FAOS Research Programme, RA-6. Under peer review; arXiv preprint posts at release.}
}
RA-12arXiv-ready
Entropy-Guided Ontology Design
Thanh Luong Tuan, Abhijit Sanyal
A design-time method for predicting whether an ontology will earn its grounding lift — structural entropy of the ontology, measured before any agent experiments, predicts downstream agent performance at Spearman r = 0.811 across fifteen industry-model cells. The interaction layer does most of the predictive work.
@article{luong2026entropy,
title = {Entropy-Guided Ontology Design},
author = {Luong Tuan, Thanh and Sanyal, Abhijit},
year = {2026},
journal = {Preprint pending arXiv submission},
note = {FAOS Research Programme, RA-12. Citation audit available at \url{https://github.com/frank-luongt/faos-research/tree/main/RA-12}; paper PDF posts at arXiv submission.}
}
In the pipeline
Active tracks. ETAs are working dates and may shift as verification continues.
RA-15Ready to post
Contextuality-Guided Orchestration
Negative-results method paper · arXiv-ready, posting soon
RA-11Date TBD
Quantum-Inspired Context Engineering
v1.0 · post-empirical findings, arXiv timing under review
Next paper in flight
Posted when it's ready.
The FAOS Research Programme
FAOS is built as Customer Zero — one founder, fifty agents, twenty-five industries — and the research programme runs on the same stack we ship to customers. Every paper has a working implementation. Every method gets tested against real enterprise workloads before it shows up here.
The programme covers four tracks: neurosymbolic enterprise AI, agent simulation and verification, ontology and context engineering, and multi-agent coordination. Papers are written for the practitioner who needs to deploy this work — not just the reviewer who needs to evaluate it.
Why we work this way is in the manifesto. The short version: systems that learn from every engagement compound. Tools that don't, commoditize.
Want to apply this work?
Most of what we publish ships first inside FAOS. Book a workshop to see how it applies to your enterprise.
