Build the intelligence
- Curated data products and semantic context
- Knowledge bases and retrieval
- Websites, portals, APIs, and reports
- Assistants, copilots, agents, and workflows
The DataKnobs platform architecture
KREATE. KONTROLS. KNOBS.
Three responsibilities for production AI: create the capability, govern how it operates, and expose the choices that must be measured, tuned, approved, and improved over time.
The intelligent trio
Separate creation, control, and adaptation.
This separation makes the system easier to reason about, govern, experiment with, and reuse across products and teams.
Govern the system
- Identity and access
- Privacy and source boundaries
- Lineage and audit evidence
- Policies, approvals, quality gates, and compliance
Make behavior adjustable
- Data selection and freshness
- Model, prompt, retrieval, and tools
- Thresholds, routing, and autonomy
- Cost limits, fallbacks, rollout, and rollback
Knobs in practice
Move consequential choices out of hidden code.
A knob is not only a model hyperparameter. It is any adjustable decision whose effect on quality, cost, risk, latency, or user outcome should be observable.
Context knobs
Source selection, lookback period, filters, chunking, retrieval depth, freshness, feature definitions, and inclusion rules.
Intelligence knobs
Model, prompt, examples, temperature, reasoning policy, tool selection, memory, confidence threshold, and workflow branch.
Action & risk knobs
Agent autonomy, approval thresholds, routing, spending limits, escalation rules, fallback models, sampling, and rollback policy.
Operating loop
Define → run → evaluate → change → learn.
Experimentation is how teams learn which knob settings work. Governance determines which changes are allowed to move into production.
1
Define
Choose outcome metrics, constraints, and adjustable knobs.
2
Run
Execute the workflow against representative production conditions.
3
Evaluate
Compare quality, cost, latency, safety, and business outcomes.
4
Promote
Approve a setting, stage a rollout, or keep a human checkpoint.
5
Monitor
Watch drift and repeat when context or business conditions change.
Examples
The same trio works across very different domains.
Complaint operations
KREATE structures interactions, KONTROLS protects evidence and routing, and KNOBS tune thresholds, taxonomies, and human review.
Earnings intelligence
KREATE turns transcripts and metrics into a data product, KONTROLS preserves provenance, and KNOBS tune scoring, prompts, and ranking.
Industrial AI
KREATE builds health signals, KONTROLS defines validation gates, and KNOBS tune detection thresholds, warning horizons, and actions.
AI websites
KREATE generates experiences, KONTROLS checks SEO/accessibility/policy, and KNOBS tune content, experiments, routing, and conversion workflows.
FAQ
Platform questions.
What are KREATE, KONTROLS, and KNOBS?
They are the three core responsibilities in the DataKnobs platform: KREATE builds reusable intelligence, KONTROLS governs boundaries and evidence, and KNOBS exposes adjustable decisions for evaluation and optimization.
Is AB experimentation a fourth pillar?
No. Experimentation is a capability used to compare knob settings, prompts, models, workflows, or experiences. It supports KNOBS and the broader platform rather than replacing the three-pillar model.
What can be a knob?
Any consequential adjustable choice can be a knob, including data source, retrieval policy, model, prompt, threshold, routing, autonomy, cost limit, approval rule, or fallback behavior.
Design the knobs before the system becomes hard to change.
A DataKnobs workshop can map your business outcome, context, controls, and measurable levers into a practical platform adoption blueprint.
Explore the Workshop