Messy data in, trusted records out
InRecipe enriches an item master the way a skilled data steward would - reading a raw part name, resolving the manufacturer, identifying the part and its specs - but at machine scale, and with every value traceable to its source.
The autopilot does the work; the safety rails make the output something you can feed straight into a production system or a customer-facing digital service.
Priced as a replacement for data-stewardship headcount, not per-seat software.
ATLAS 90kW VSD #2Atlas Copco GA 90 VSD · 90 kW · variable speed (VSD)Manufacturer resolved, every field cited to its source. Where a raw name fits more than one model, InRecipe flags it for review instead of guessing.
The cost is everything downstream
One missing or untrusted spec rarely feels critical. But the effects cascade - into margin lost, revenue delayed, and compliance exposure. These are the costs industrial companies name themselves.
Options sold at a loss no one can trace
When customers upgrade, costs blur together in the ERP - so no one can see which options make money and which sell at a loss. Clean cost data makes the loss-makers visible.
Devices that can't be configured to sell
Product data lands weeks late, so new models can't be added to the sales configurator in time for launch. Late data is late revenue.
Tens of thousands in customs exposure
Wrong or missing customs classification on cross-border shipments exposes the business to back-duties and penalties - and the importer, not the supplier, is legally liable.
Under-invoicing on stale price lists
When price lists aren't updated, contracts bill customers too little while parts are bought at today's higher cost - margin lost on every job.
For companies whose data runs the business
Manufacturers, distributors, and service businesses where incomplete or inconsistent equipment and parts data quietly blocks pricing, sales, sourcing, and compliance.
Item masters that feed the whole value chain
One missing part stalls sourcing, breaks the configurator, and delays service. Clean data unblocks design through after-sales at once.
Catalogs too large to maintain by hand
Thousands of items across many principals, each needing specs, classification, and customs codes. Manual stewardship can't keep pace.
Fleets priced on what's installed
Maintenance pricing, contracts, and planning all depend on knowing the real equipment base - accurately, at scale.
The common thread: the data is an asset other systems depend on - and getting it wrong has a direct business cost.
From a device name to dozens of structured fields
From just a device name, InRecipe identifies the manufacturer, model, category and full technical specs - dozens of fields, each backed by a source you can check.
Compair L15-10A K1 screw compressorParse
Extract known patterns from the device name and model code.
Search
Look up specs from manufacturer websites and datasheets.
Enrich
Fill gaps from verified sources, standardize units, and flag anything it can't confirm.
All powered by AI - including rule generation.
Built like a kitchen
Reusable techniques combine into recipes; recipes are collected into a cookbook for each industry; and the kitchen that runs them never changes. That's what lets a new industry ship in days, not months.
The cookbook
Manufacturer catalogs, vocabulary and validation rules for a whole industry. Grows denser and more valuable with every customer.
The recipe
Decides which techniques run, in what order, on which data, and for what business purpose.
The technique
Does one thing well: resolve a manufacturer, extract a spec, classify a customs code. Reused across every industry, unchanged.
The evidence contract
A one-shot AI gives you a plausible answer. InRecipe gives you an answer you can prove - or an honest “no source found.” Every value is checked by code before it reaches your database.
The AI never produces a bare value
It produces a value and a citation - which source, which field, which catalog entry.
The system verifies the citation
The claimed source is checked against the curated catalog and the input data. If any check fails, the value is rejected.
Only verified values enter the database
What can't be verified is flagged or left honestly empty - never silently guessed.
Taken from a manufacturer source with a citation that points back to it. Highest trust.
A domain-bounded estimate when no source is found - capped at lower confidence, flagged for review, never disguised as fact.
Researched and nothing trustworthy turned up. An honest blank - distinct from “didn't look.”
Five stops, no shortcuts
Every record takes the same auditable path. What runs is declared up front; what comes out is checked by the system. Nothing reaches the database on the model's word alone.
Customer file
The item master as it is - messy names, mixed languages, gaps.
The order
Declares which skills run and which fields are in scope - the surface you sign off on.
The work
Each skill reads sources - catalogs, spec sheets, the web - and emits a value with its evidence.
The gate
Checks every claim: real source, allowed evidence, confidence within ceiling. Any check fails - rejected.
Graded record
Every field carries its source, date, and confidence - an audit trail, not just a value.
Smarter across customers without sharing your data
Every value InRecipe touches has an owner. We sort them into three tiers by who that owner is - and that ownership, not a policy you have to trust us on, decides what's shared and what stays yours.
Manufacturer facts
Model-number grammars · published power, pressure and flow · manufacturer-declared country of origin.
Regulatory & supplier facts
Customs / HS codes · supplier SKUs, item names, publicly available lead times and list prices.
Your operational reality
Service & change intervals · work-order history · contract terms · your in-house names and categories.
Curator-reviewed before anything is shared. The pattern travels across your vertical; your records never do.
Industrial air systems: 2,500 devices enriched in under 3 hours
The challenge
A service business priced its maintenance contracts on the equipment each customer had installed. But the fleet records were patchy and inconsistent - manufacturer, power rating, type classification largely missing - so pricing leaned on guesswork, and margin leaked on every contract.
What we deployed
InRecipe in numbers
A shelf of reusable skills, several domain packs, and a growing base of curated manufacturer knowledge - the assets every new engagement builds on.
Domain packs
Each a self-contained vertical: catalogs, vocabulary, validation rules. Industrial air, marine, mining spares, industrial valves.
Skills built
Reusable units of capability - 10 distinct types, composed differently per pack and shared across them.
Curated manufacturers & distributors
Verified commercial entities - the closed-world catalogs the evidence contract checks against.
Skills are shared across packs; only the content differs. Behind it: ~30 documented failure modes and the test fixtures that encode them - knowledge a competitor must relearn engagement by engagement.
What changes in your register
Today
With InRecipe
Two ways to run it
Start with a one-off cleanup. Or embed enrichment into your data platform so every new device enters the register clean.
InRecipe Copilot
Upload equipment registers, enrich and standardize, export clean data.
InRecipe Autopilot
Integrate enrichment directly into your existing infrastructure. New devices entering the register are automatically enriched, categorized, and linked.
Built for the security questionnaire
Where your data goes, which models touch it, and how the architecture maps to where EU AI regulation is heading.
Local processing, isolated tenants
Processing runs in Data Design-controlled environments. Records, outputs, and history stay in your engagement tenant.
A governed model layer
Calls go to commercial frontier-model APIs (e.g. Anthropic's Claude). Inputs and outputs are not used for training. Pipelines are model-agnostic.
AI Act - shaped by architecture
Human oversight, transparency, and audit trails are core to the product. Minimal personal data keeps GDPR exposure narrow.
See it on your own data, not a demo
A short, low-risk path from a sample of your register to validated, evidence-graded output - then continued use priced against the value delivered.
Copiloting
One business case, 1–2 manufacturers or distributors in scope, and a 50–100 record sample of your equipment or parts data.
Data Design bears the LLM costs, analyzes the data, and builds initial catalogs & recipes. You see real, evidence-graded output - not a demo on someone else's data.
InRecipe Copilot
Deep analysis and building curated manufacturer catalogs, with enrichment tailored to your specific needs.
Most device or spare-parts data enriched with high confidence.
Priced against the volume of data.
InRecipe Autopilot
Fully automatic enrichment for new records - seamless integration and maintenance-free data quality.
Monthly recurring fee, tiered on data volumes.
