AI vehicle diagnostics · Prototype in field validation

Find the fault. Prove the root cause.

ADIE is an AI diagnostic assistant that ranks the likely faults behind a vehicle's symptoms and shows the evidence for every conclusion — sensor readings, CAN-bus data, direct electrical measurements, and the vehicle's own history, kept in an auditable evidence ledger. No parts cannon. No black-box guess.

See what works today The three capabilities

3 core capabilities 2-layer diagnostics: bench + CAN 100+ verified cases = proof target

Technician running laptop diagnostics over an open engine bay

ADIE ranked diagnosis Example case

2013 Hyundai Elantra GLS — engine tick

1 · Timing-chain tensioner wearLeading

Tick loudest at the timing cover; worst on cold start, fades as oil pressure builds.

2 · Hydraulic lifter tickPossible

Tick at half engine speed from the valve cover; check oil level and condition first.

3 · Exhaust manifold leakPossible

Ticking that quiets as the manifold heats and expands; soot traces at the flange.

Illustrative output from ADIE's field-validation case list. A 20-minute stethoscope check — where the noise is, and when it happens — separates these three before any parts are ordered.

Honest status

What's working today — and what isn't yet

ADIE is a working prototype, not a finished product. Everything on this page is labelled by where it actually stands, because a diagnostic tool that overstates its evidence isn't worth trusting — and neither is the company building it.

Working in lab tests Verified

Injector & ignition bench tester. An Arduino-based rig drives each injector and measures current draw through a precision shunt, checks coil resistance, and confirms actual spark per cylinder — output streamed as data lines ADIE ingests directly.
Failure-data pipeline. Pulls real NHTSA complaint and recall data by vehicle and mines it into per-component failure-mileage curves — tested end-to-end on the 2013 Elantra (1,330 complaints analysed).
Remaining-life estimator (prototype). Combines those population failure curves with a vehicle's mileage, region, and maintenance history to flag components entering their high-risk window.
Verified case logger. Every field case records ADIE's call, the actual fault found, and whether the fix worked — the raw material for a measurable accuracy record.
Synthetic fleet testbed. Lifelike generated vehicle histories, sampled from the real NHTSA failure distributions, used to stress-test the math. Watermarked and kept strictly separate from real cases — synthetic data is for testing, never for claims.

Still under development In development

Real-car CAN integration. The deep-data path — a Wi-Fi OBD-II dongle streaming live bus data to ADIE — is designed and bench-staged. Live capture on a real vehicle, including CAN bitrate validation, is the current critical path.
Predictive maintenance alerts. The estimator catches nearly everything that fails in testing (92% recall) but over-warns (11% precision). Calibrating it against real repair outcomes — not simulations — is what comes next.
Root-cause ranking at scale. Ranking works on the systems the bench rig measures directly. Extending it across full-vehicle systems depends on the verified case library growing.
Vehicle Memory & Trust Score. A lifelong record per vehicle, and a score buyers and sellers can rely on. Designed; needs real vehicles flowing through the pipeline first.
Fleet intelligence. Cross-vehicle pattern detection needs a fleet's worth of verified data. It's the last phase of the roadmap, deliberately.

Why publish the gaps? ADIE's whole premise is evidence over assertion. The validation plan below is public, the accuracy targets are public, and every case ADIE gets wrong stays in the log. That's the standard customers, investors, and funding organizations should hold us to.

What ADIE does

Three capabilities, one evidence trail

Diagnosis, prediction, and analysis aren't three products — they're one loop. Analysis feeds diagnosis, diagnosis outcomes teach prediction, and every step writes to the same ledger.

AI-powered diagnostics

Lab-tested

Give ADIE the symptoms, the fault codes, and the readings — it returns a ranked list of likely faults and investigates down to the root cause, not just the code that lit the dash.

Ranked, not binary. Every candidate fault carries its supporting and contradicting evidence, so a technician can test in order of likelihood instead of replacing in order of price.
Root cause over symptom. A misfire code is a symptom. ADIE separates weak coil, failing injector, and mechanical causes — including the weak-but-still-firing coil that crankshaft-based misfire detection misses entirely.
Direct measurement where it matters. The bench rig measures what the car's computer can only infer: coil resistance, per-injector current draw, confirmed spark per cylinder.

Gloved hands working on an engine with a wrench

Predictive maintenance

In development

The goal: flag a developing problem while it's still a cheap fix — before the component fails, strands the driver, or takes neighbouring parts with it.

Failure curves from real data. Per-component failure-mileage distributions built from 1,300+ owner complaints per model, so "watch this at 180,000 km" comes from evidence, not a generic service interval.
Remaining useful life. Population curves adjusted by the individual vehicle — its mileage, salt-belt exposure, and maintenance history.
Failure weather forecast. The roadmap view: a forward risk outlook per vehicle, the way a forecast shows storms forming days out — honest about uncertainty, updated as new data lands.

Technician reviewing live vehicle data graphs on a rugged tablet under a raised car

Intelligent vehicle analysis

In development

A modern vehicle is dozens of computers talking over a CAN bus. ADIE brings that conversation together with sensor data and the vehicle's history into one analysable record.

ECU + CAN-bus data. Fault codes (DTCs), misfire counts, fuel trims, RPM, load, and injector pulse width — what the ECU sees, read directly.
Physical layer cross-check. Bench measurements verify the ECU's interpretation. When the two disagree, that's a finding in itself — and it goes in the ledger.
History that compounds. Connected-car state (odometer, battery, tire pressures, codes), service records, and every past ADIE case build a Vehicle Memory that makes each new diagnosis sharper than the last.

Technician holding a diagnostic tablet beside a vehicle in a repair shop

How it works

Every diagnosis leaves a trail

The evidence ledger is the difference between an opinion and a diagnosis. Each step is recorded, so any conclusion ADIE reaches can be audited back to the readings that produced it.

STEP 1

Capture the signals

Symptoms in the driver's words, fault codes and live data from the CAN bus, and — where the bench rig applies — direct electrical measurements.

STEP 2

Rank the faults

Candidate faults are scored against the evidence and against failure patterns for that exact make, model, and mileage — most likely first.

STEP 3

Show the evidence

Each ranking links to the readings behind it in the evidence ledger. A shop can show the customer exactly why a part is being recommended.

STEP 4

Record the outcome

The actual fault and whether the fix worked are logged against the prediction. Right or wrong, the case makes the next diagnosis better.

Two layers beat one: what the bus can't tell you

CAN data is the ECU's interpretation of the engine — powerful, but indirect. ADIE's second layer measures the physical reality directly and cross-checks the two. That cross-check is where cheap, confident diagnoses come from.

The classic miss: a coil that's weak but still firing. The ECU infers misfires from crankshaft speed wobble and may never flag it. The bench rig sees it immediately — low spark energy and off-spec resistance, measured, not inferred.

Signal	Via CAN / ECU	ADIE bench layer
Misfire counts, fault codes (P0300–P0312)	Yes	—
Fuel trims, RPM, engine load	Yes	—
Injector pulse width (many models)	Yes	Cross-check
Actual injector current draw	No	Measured
Coil resistance	No	Measured
Direct spark confirmation, per cylinder	No	Measured

Validation

Proof is a ladder, and we're climbing it in public

Ten documented cases beat a hundred undocumented ones. Every ADIE field case records the same three things — ADIE's call, the actual fault, and whether the fix worked — and the strongest test is head-to-head: ADIE's diagnosis against a shop's, on the same car.

Vehicles — it functions

Ten real cars through the full pipeline end-to-end proves the system works outside the lab. This is the rung currently being climbed: the data stack is tested, the bench rig works, real-car capture is next.

30–50

Repair cases — the concept holds

Thirty to fifty cases with recorded outcomes prove the concept: a real hit rate, measured against what was actually wrong — not against what a simulation says should be wrong.

100+

Documented cases — it works

A hundred-plus cases tightens the confidence interval enough for partners, shops, and funders to rely on. Fifty head-to-head wins against shop diagnoses is a story nobody can dismiss.

1,330

NHTSA complaints analysed for one test vehicle (2013 Elantra) to build its failure curves

92%

Recall of the remaining-life flag in synthetic-fleet testing — it catches nearly everything that fails

11%

Precision in the same test — it over-warns today. Published here because calibrating it on real cases is the work

Facts logged per case, every time: ADIE's call · the actual fault · whether the fix worked

Roadmap

Built in phases, data first

The sequence is deliberate: real-car data is the critical path, because every intelligent feature downstream is only as good as the verified cases underneath it.

Phase 1

Real-car data pipeline

Live CAN capture from real vehicles via Wi-Fi OBD-II, alongside the bench rig's direct measurements and always-on state from connected-car services.

Wi-Fi OBD-II dongle · Bench rig · OEM connected-car APIs · NHTSA knowledge base

In progress

Phase 2

Vehicle Memory & outcome capture

A lifelong record per vehicle — every reading, diagnosis, repair, and outcome — and the verified case library that turns individual repairs into Repair Success Intelligence: which fixes actually work, on which cars, at what mileage.

Vehicle Memory · Repair Success Intelligence · Verified case log

Phase 3

Single-vehicle intelligence

Vehicle “Aging DNA” — how this specific car is ageing against its population — with a failure forecast the owner can act on, and risk-based maintenance that replaces fixed intervals with the vehicle's real condition.

Aging DNA · Failure Weather Forecast · Risk-Based Maintenance

Planned

Phase 4

Causal & economic layers

The Hidden Failure Network — faults that hide behind other faults — plus a remaining-useful-life and market-value engine, and a Vehicle Trust Score grounded in the car's verified history.

Hidden Failure Network · RUL / value engine · Vehicle Trust Score

Planned

Phase 5

Fleet Swarm Intelligence

Patterns no single vehicle can show: emerging failures across a fleet or a model line, detected from thousands of vehicles at once via fleet telematics — the early-warning system for the faults nobody has named yet.

Fleet telematics (Geotab · Samsara) · Cross-vehicle pattern detection

Planned

Who it's for

One tool, two conversations

Technician reviewing diagnostic data on a tablet beside a car with its hood open

For drivers & repair shops

A scary symptom — an engine tick, a misfire, a warning light — usually gets priced as the worst case. ADIE exists to find out what's actually wrong before money changes hands.

Drivers: a ranked, evidence-backed second opinion before you authorise a repair — or before you walk away from a good car priced like a dead one.
Shops: faster triage on the hard ones, and an evidence trail you can show the customer instead of asking them to take your word for it.
Sellers & buyers: field-validation cases start with exactly the cars everyone else misprices — cheap, newer cars with frightening symptoms.

Put a vehicle forward

For investors & funding organizations

The asset being built isn't an app — it's a verified dataset of diagnoses and outcomes that compounds with every case, in a market where diagnosis is still mostly experience and guesswork.

A data moat with a paper trail: every case logs the prediction, the real fault, and the fix outcome. Accuracy is a number, not a pitch.
Honest metrics, published: 92% recall / 11% precision on the current estimator, stated plainly — the calibration gap is the roadmap, not a footnote.
Phased, capital-efficient plan: field validation starts with a $15-class OBD dongle and a case log, not a hardware fleet. Spend follows proof.
Multiple expansion paths: shop tooling, fleet intelligence, vehicle trust scoring, and remaining-life valuation all build on the same case library.

Request the case for support

Start a case

Tell us about the vehicle — or the opportunity

Have a car with a symptom nobody's pinned down? A shop that wants sharper triage? A funding or partnership conversation? Fill in the brief and download it — it formats everything ADIE needs to open a case, ready to send.

Field-validation vehicles wanted

Newer cars with scary symptoms and scary quotes are exactly what the validation ladder runs on. Diagnosis first, verdict second.

Shops

Early shop partners get the ranked-diagnosis workflow and a say in how the evidence report reads to customers.

Investors & funders

Ask for the validation plan and the case-log format. The numbers on this page are the numbers in the log.