Behind Engine Match

How the advisor works — and what a real industry model is made of.

A trustworthy automotive AI isn't a clever chatbot. It's three conceptual zones — trusted knowledge, an automotive intelligence layer, and the experience on top — each held to a standard, with a human at the end of the phone.

The LLM is not the knowledge base. It is the reasoning engine over a governed, continuously expanding body of automotive knowledge.

The stack — seven layers, three zones

Zone 3 · Consumer & Expert Experience

07AI advisor + human expertThe phone

Zone 2 · Automotive Intelligence + LLM

06Automotive intelligence layerGrounded LLM

Zone 1 · Trusted Knowledge

05Trusted external knowledgeVetted sources 04Installation & field knowledgeService records 03Manufacturer & performance dataDyno cell 02Fitment data — ACESACES 4.2 01Product data — PIESPIES 7.2

Layers 06–07 are where the conversation happens — everything below is their evidence.

Zone 1

Trusted Knowledge

The evidence the advisor can draw from

Not all data is equal. The advisor prioritizes manufacturer-validated product and fitment data, then layers in trusted industry and real-world sources for additional context. Every answer retains a connection to its source.

01PIES 7.2

Product data — PIES

Every part is a machine-readable record, not a paragraph

The Product Information Exchange Standard defines how a part describes itself: part number, brand AAIA ID, PCdb part terminology, dimensions, weights, country of origin, pricing and lifecycle status. When product data is structured this way, an advisor can compare, filter and cite it with confidence — a paragraph of marketing copy can't be verified, a record can.

In this demo

This demo: sample PIES records for the crate-engine catalog.

In production

Production: the live published PIES feed, refreshed on a schedule.

02ACES 4.2

Fitment data — ACES

Fitment is structured, not prose

The Aftermarket Catalog Exchange Standard answers the question 'does it fit?' with data: year/make/model/engine combinations keyed to BaseVehicleIDs, plus qualifiers for transmission, emissions, fuel system and drive. A fitment answer you can look up beats a fitment answer you have to trust.

In this demo

This demo: sample BaseVehicleIDs and fitment qualifiers per engine.

In production

Production: live ACES fitment against the full PCdb/VCdb vehicle table.

03Dyno cell

Manufacturer & performance data

The written number is the ground truth

Horsepower and torque claims mean something only when they're produced by a documented method. Every BluePrint engine is dyno-tested before it leaves Kearney, and rotating assemblies are balanced within two grams. Manufacturer specifications, dyno methodology, tolerances and test conditions are first-class data — the layer that keeps every other layer honest.

In this demo

This demo: verified claims (dyno-tested, two-gram balance) held as constants in the advisor's prompt.

In production

Production: per-engine dyno records and specs joined to the product record.

04Service records

Installation & field knowledge

What installers and technicians actually learn

Installation documentation, service bulletins and field experience hold knowledge no catalog has: damper and timing-tab compatibility, flexplate requirements, clearance on stock hoods, what a warranty claim actually looked like. Captured and governed, this becomes institutional memory instead of tribal knowledge.

In this demo

This demo: a curated, dated snapshot of owner-forum discussion, labeled as anecdote.

In production

Production: periodically refreshed, source-attributed field data with recency dates.

05Vetted sources

Trusted external knowledge

Selected, weighted, and never blended with the record

Selected automotive websites, technical publications, trusted forums and communities, regulatory and government sources, and other vetted industry sources extend the model's reach. But external knowledge is weighted differently from manufacturer-validated data — the system tracks provenance on every fact, and anecdote can never override an official specification.

In this demo

This demo: clearly labeled community excerpts, attributed to source and year.

In production

Production: a governed registry of vetted sources with per-source trust weights.

Zone 2

Automotive Intelligence + LLM

Grounded LLM + semantic automotive data + source hierarchy

This is where the LLM takes the customer's question and reasons across the trusted knowledge: understanding relationships, retrieving the right evidence, resolving conflicts where possible, and producing a recommendation with evidence and confidence attached. Crucially, the data is not fed into the model permanently — production architecture is retrieve-then-reason: question → retrieve relevant trusted data → provide it as context → reason → answer. The knowledge can grow and change every day without retraining the LLM.

06Grounded LLM

Automotive intelligence layer

A reasoning engine over governed knowledge — not a knowledge base

The intelligence layer is constrained by strict rules: recommend only catalog products, always cite the PIES part number and ACES BaseVehicleID, weigh sources by their provenance, say 'I don't know' when the evidence isn't there, and hand off to a human rather than improvise. The defensible asset is the trusted, structured knowledge the model reasons over — anyone can plug a chatbot into a website.

In this demo

This demo: a system prompt enforcing catalog-only, referenceable answers.

In production

Production: retrieval over the live feed, with evals against known fitment cases.

Zone 3

Consumer & Expert Experience

Where the conversation happens

The top of the stack is the experience itself: a recommendation, the supporting evidence, a stated confidence level, clarifying questions when the question is ambiguous — and human escalation whenever the stakes or uncertainty call for it. The AI is the entry point, never the last word.

07The phone

AI advisor + human expert

Recommendation, evidence, confidence — then a person

Every answer cites its sources and states its confidence. When a build is unusual, a photo is ambiguous, or the stakes are high, the easiest step in the whole experience is the handoff to an engine specialist. A well-built model knows where its knowledge ends.

In this demo

This demo: sales and tech-support phone numbers surfaced in every answer.

In production

Production: conversation handoff with full context passed to the specialist.

See it in action

Try the advisor — then talk to a person.

Every Engine Match answer cites a part number you can verify, and every answer ends one step away from a specialist on (308) 236-1010.

Open Engine Match Contact us