Pike Labs
Cognitive neurosymbolic AI
How a Pike-based AI system is put together

First, codify how your business works. Then let Pike broker the models against it.

Codifying means writing down, as executable knowledge, what your organization is trying to achieve, how it goes about it, what it watches, and the rules it works by, including the ways of working that are particular to you. Once that exists, Pike can stand between it and any model: asking narrow questions, checking the answers, and keeping the intent, the memory, and the explanation on your side.

Your enterprise Operators & domain experts author and bound the charter, approve, and receive the why HUMAN-CENTERED · ALWAYS IN THE LOOP Systems of record EHR & claims · telemetry · ERP documents · legacy BI STAYS WHERE IT IS · READ, NOT MOVED Events & signals devices · transactions · messages and the passage of time REAL-TIME OR TURN-BASED People & agents, observed the actions Pike explains against your model of how work should go INTENT · ERROR · GAP IN KNOWLEDGE IN → charter · data · events · actions OUT ← plans · alerts · decisions the right information, and the why Pike · the cognitive core Codified domain — your business as executable knowledge Ontology concepts, roles, vocabulary, relationships Intent model goals → plans → actions, with the bounds on each World-state model abstract states inferred from concrete signals Policy & custom capability rules, statutes, your playbook, your own models and analytics Engineered by your experts, drafted with AI assistance, always under human review. Cognitive engines — the algorithms never change Plan & execute Interpret intent Track the world Triage information Every decision carries its explanation chain — the reasoning, not a story told after. The broker — Pike stands between your domain and the models when a model is asked — only where symbols run out what it sees — a bounded, pseudonymous slice, never the whole which model answers — frontier, small, private, or your own whether the answer holds — checked against the structures before it binds what it may become — a value in the model, never an action on its own Models advise. The structures decide. EVERY CALL LOGGED · EVERY ANSWER TRACEABLE TO A GOAL BOUNDED QUESTION JUDGMENT, CHECKED Models — the neural half LLM · frontier SLM · private custom & domain analytics · BI INTERCHANGEABLE · STATELESS · YOURS TO SWAP What it yields 01Explainable by construction the explanation chain is the reasoning, not a rationalization after the fact 02Purpose that persists goals live in the intent model and outlast any single prompt, session, or model 03Memory beyond the window the world lives in the structures; the context window is a slice, not the ceiling 04Coordinated by design agents share one intent model; a subtree can be handed to a bounded sub-agent 05Autonomy with a dial the charter says how far it may go alone; operators set the dial per situation 06Safe by structure a model can only fill what the structures admit; safety is a property, not a filter 07Your alpha, protected your judgment and playbook stay in your structures, on your infrastructure; models see bounded slices and are swappable
Cognitive neurosymbolic AI — the symbolic half holds the intent; the neural half fills what resists symbols.
What the picture says

The knowledge is the product. The model is a component.

Most enterprise AI today puts the model at the centre and pours the business into it: documents into a retrieval index, rules into a system prompt, the company's judgment into whatever a fine-tune can absorb. The result is capable, opaque, and forgetful, and the part that made the business distinctive now lives inside a vendor's weights.

A Pike-based system inverts that. The business is written down first, as executable knowledge: what the organization is trying to achieve, the ways it pursues those aims, what counts as the world being in a given state, and the rules and playbook that bound all of it. Pike's engines run over that knowledge. They plan, they interpret what people and systems are doing, they keep track of the world, and they decide what deserves attention. Only where symbols run out does Pike turn to a model, and when it does, it asks a narrow question, with a bounded slice of context, of the model best suited to that question, and checks the answer before it is allowed to matter.

01

Explainable by construction

Every decision arrives with the chain of goals and plans that produced it. The explanation is the reasoning, because the reasoning is the architecture.

02

Purpose that persists

Goals are held in the intent model, not in a prompt. They survive the length of the work, a change of model, and a change of shift.

03

Memory beyond the window

The state of the world lives in the structures and in your systems of record. A model sees a slice when it is asked; nothing degrades as the work accumulates.

04

Coordinated by design

Agents share one model of the domain. A part of the work can be handed to a bounded sub-agent with its own slice of the structures, and it reports back in the same terms.

05

Autonomy with a dial

The charter states what may be done unattended and what needs a person. The dial moves per situation, per customer, and per regulator, without retraining anything.

06

Safe by structure

A model can only fill in what the structures admit. There is no action it can take that the knowledge does not already name and bound.

07

Your alpha, protected

The edge a company has that no general model has is its own judgment: how it decides, what it watches, what it will and will not do. In a Pike-based system that edge lives in your structures, on your infrastructure. Models see bounded, pseudonymous slices, learn nothing durable from you, and can be swapped as the market moves.

Why the model gets better, not smaller

Brokering magnifies what a model is worth.

One core, many worlds

Business domains where the Pike neurosymbolic architecture is being applied.

The engines never change. Each domain fills the codified structures differently and brokers different models. Three that are in progress:

Cardiac rhythm care

Remote monitoring and electrophysiology

Codified
rhythm findings, device and signal quality, care pathways, and the reimbursement rules that follow them
Brokered
the ECG classification model, and a language model that summarizes for the clinician
Result
triage across a population that explains itself, and billing that is defensible

Mass transit operations

Regional transit agencies

Codified
fleet and vehicle state, service plans, operating rules, and incident playbooks
Brokered
telemetry anomaly models, and language models over rider and crew communications
Result
the right alert to the right desk, with the reason attached

Physician billing compliance

Surviving a CMS audit

Codified
statutes, recent rulings, audit practice, and the claim patterns auditors question
Brokered
a language model reading clinical notes and claims
Result
pre-audit jeopardy flagged early, and a claim-by-claim explanation chain as the defense file, feeding counsel rather than replacing it