Engineering Intelligence

A general assistant answers. An engineering copilot has to be right.

The difference is not the size of the model. It is what the model is permitted to decide, what it must cite, and what happens when it does not have enough information.

The distinction

Four differences that matter in hardware.

Dimension A general AI assistant Suyantra AI Copilot
Source of truth Whatever the model recalls, expressed fluently. Retrieved evidence, cited to the document section it came from.
Who decides The model produces the conclusion. Deterministic rules produce the verdict; the model explains it.
Missing information Filled with a plausible typical value. Check returns Blocked with the missing input named.
Consequences of change Not modelled — each question is answered fresh. Computed from a recorded dependency graph.
The provenance model

Every value knows where it came from.

This is the single idea the whole product rests on. A number without a provenance tag is a number you have to re-derive yourself, which means the tool saved you nothing.

Stated

Given to us

The value appears in the requirement. Suyantra treats it as fixed and will flag a conflict rather than quietly relax it.

Square marker
Inferred

Derived from stated values

Computed, with the computation shown. If an input changes, everything inferred from it is recomputed.

Circle marker
Assumed

Not stated — declared anyway

Where an assumption is unavoidable, it is recorded as one, attributed to the product, and open to being overturned.

Diamond marker
Verified

Checked against evidence

A rule ran, had all its inputs, and passed. Verification is cleared automatically if a dependency later moves.

Ringed marker
Colour is never the only signal. Each provenance tag carries a distinct marker shape and a text label, so the distinction survives greyscale printing and colour vision differences.
Confidence

A confidence figure without a basis is decoration.

Suyantra states confidence alongside the reason for it — how many parameters were verified, which inputs were missing, whether the check was deterministic or interpretive.

HighDETERMINISTIC
AssessmentHigh
Basis: a rating rule with all inputs present, evaluated against cited datasheet parameters.
MediumPARTIAL INPUTS
AssessmentMedium
Basis: 2 parameters verified, 1 thermal input missing. The gap is named rather than smoothed over.
LowSCOPE UNKNOWN
AssessmentLow
Basis: the requirement is silent on this area, so the finding is a prompt for clarification, not a conclusion.
Known limits

What Suyantra does not do.

Stating this plainly is part of the product. An engineering tool that is vague about its boundary is a tool you cannot rely on at the boundary.

It does not replace simulation

Rating and margin checks are not a substitute for circuit simulation, thermal modelling or EMC testing.

It does not qualify a design

Qualification, compliance testing and certification remain the responsibility of the engineering organisation.

It does not sign off

Suyantra can report that a design is not yet defensible. Declaring it finished is an engineering act performed by a person.

It is only as good as its evidence

Where a datasheet does not publish a parameter, no amount of reasoning recovers it. The check blocks instead.

It cannot see what is not recorded

A dependency that was never captured cannot appear in an impact set. The graph reflects the design as entered.

It is in active development

Scope and capability are changing. We would rather tell you what is not ready than imply it is.

Get started

Tell us where your reasoning currently lives.

Spreadsheets, review comments, someone's head — we would like to hear how your team works today.