A neuro-symbolic paradigm
Decisions a machine can prove, not just produce.
rAIson pairs a large language model (LLM) with structured preference-based argumentation compiled into propositional satisfiability (SAT) — so high-stakes decisions come out reproducible, explainable, and defensible. Built without writing a single line of code.
A different way of building AI: anyone who can describe their policy — as natural-language text or as tabular data — can take it from there to a deployed agent. No programmers, no AI engineers required. The platform is coming to the public; today, our team already builds with the same technology for individuals and companies alike.
Reads and generalises
- Detects and extracts reasoning patterns from text and tabular data
- Maps natural language to formal rules
- Generalises across cases and variants
Decides and justifies
- Deterministic, reproducible verdicts
- Every conclusion has a named-rule trace
- Milliseconds on a commodity CPU
The idea
Each paradigm has exactly what the other lacks.
Language models read fluently but answer stochastically, with no trace to audit. Computational argumentation decides deterministically (unless there is a justified dilemma) through a non-monotonic reasoning mechanism and explains every step, but cannot read a source document. rAIson joins them at the level of phases, not representations: the model encodes a policy once; the symbolic AI engine (structured argumentation compiled to SAT) then executes every decision.
Encode once · execute many.
No hallucinations at decision time
The model is consulted once per source, and only to encode rules a human in the loop then approves. It never decides — so a hallucinated verdict is structurally impossible.
Every decision is a plain symbolic AI engine call over approved rules: deterministic, cheap, and backed by the rules that fired and the counter-rules they defeated.
Why rules, why now
The rule captures the cause. The trace captures the explanation.
A rule is one of the most natural forms for capturing the causal link between the reasons that warrant a decision and the decision itself: if these conditions hold, that conclusion follows. Every law, clinical protocol or eligibility test already reads this way.
When a system can name the cause that produced an effect, it explains itself by construction (explainable AI). The trace of which rules fired is the explanation of the reasoning process — not a narrative manufactured after the fact.
And not all rule systems are equal. Traditional rule-based AI is monotonic: a derived conclusion stays valid regardless of what comes after, and the system has no mechanism for handling contradictions. Law, medicine, finance and defense work the opposite way — a conclusion holds only as long as no stronger argument defeats it, and becomes valid again if that defeater is itself defeated. rAIson's engine is non-monotonic by construction.
One of the persistent difficulties of symbolic AI has been scaling up: writing and maintaining the rules is human-intensive, which limited how far the approach could go. The neuro-symbolic pairing eases this: Claude drafts the encoding from natural-language policy or tabular data, the owner approves it, updates flow the same way.
The cause is the rule. The rule is the trace. The trace is the explanation.
Trace, not narrative
When a language model is asked to explain a decision, it generates a plausible-sounding story — sometimes faithful, sometimes invented, never falsifiable from outside the model.
An argumentation trace is the opposite: a sequence of named rules anyone can re-run, dispute, or amend. The why is not authored by the reasoner — it is the reasoner.
A regulator, a clinician, or an auditor can replay the same trace and reach the same verdict, without ever calling the system that produced it. The reasoning travels with the decision; the decision travels with its proof.
Built for high stakes
Four properties law, health, finance and defense need at once.
No single paradigm has historically delivered all four. The pairing does.
Reproducible
The same inputs produce the same outputs every time. Stochastic systems fail on audit.
Explainable
Every decision arrives with an inspectable justification — a rule trace, not a vibe.
Defensible
Decisions withstand a regulator, a court, a prescriber, a journalist.
Low cost
Millions of decisions a day on commodity CPUs — no GPU at decision time.
Four ways to build
One platform, four routes to a deployed agent.
Every route ends the same way: a reasoning agent live on rAIson, reachable through one API.
Dialogue
Build directly on the platform through a natural-language dialogue. No upload, no code — you just answer the platform's requests and the PAF code is automatically generated for you.
Route 02From text
Upload a plain-language specification — a policy or scenario describing the decisions the agent should make. Claude drafts the encoding; you validate it.
Route 03From data
Upload tabular / CSV data. Claude drafts the encoding from your columns; you validate it.
Route 04Done for you
The Argument Theory team builds the agent for your domain using the same infrastructure.
At decision time
Where it applies
- Regulatory compliance
- Medical-diagnosis support
- Judicial decision support
- Risk & threat assessment
- Data-access control
- Intelligent trading
- Personal assistants
Real-world domains
Wherever a policy decides what should happen.
If a rulebook prescribes what to do under competing conditions — modal verbs, defaults and exceptions, explicit overrides — it can become an executable, auditable decision service.
Argument Theory
We offer rAIson.
Argument Theory is a deep-tech startup built on decades of internationally recognized research in computational argumentation, autonomous agents, multi-agent systems and symbolic AI more generally. We offer rAIson — our platform built on a neuro-symbolic architecture that pairs large language models (Claude) with argumentation-based reasoning encoded in propositional satisfiability (SAT).
It is how we turn the academic fields of argumentation and SAT based reasoning into production decision systems: automated decisions an organisation can stand behind — auditable, revisable, and grounded in a transparent logic rather than an opaque model.
And it changes who can build — and at what cost. Domain experts, people with innovative ideas, and companies take a policy from natural-language text or tabular data to a deployed agent without writing code or hiring AI specialists, cutting development time and cost substantially. Every verdict is deterministic and traceable, so an rAIson agent slots into an existing IT chain as a reliable component, with no added uncertainty and no risk to break the chain.
What rAIson brings
- A neuro-symbolic architecture: LLMs that read, symbolic AI that decides
- A symbolic-AI engine combining the high expressive power of a structured preference-based argumentation framework (PAF) with the computational power of SAT solving.
- Domain work in law, health, finance, defense and beyond