Find every version of the same part.
Fix what's wrong. Nothing leaves the building.
Exl Athena searches your master data the way people actually type, grades its health with AI in about a second, and surfaces the exact duplicates and gaps costing you money — using local embeddings and a local LLM, so no catalog data ever leaves your network.
in a 5,000-item sample catalog
across 6 quality rules
and score a live sample
Master data is quietly broken — and it costs real money.
The same part gets entered a dozen ways. Abbreviations and unit variants hide the duplicates. Nobody fully trusts the catalog — so teams re-buy what they already have and carry inventory they don't need.
Hidden duplicates
1,267 duplicate records across 563 groups in a single 5,000-item sample — most invisible to exact search.
Inconsistent wording
“CS” vs “carbon steel”, 3" vs 3 IN vs “3 inch”. Same item, unrecognisable to a GROUP BY.
Silent gaps
Under-specified and mis-classified records that only surface when a purchase or audit goes wrong.
Athena finds it, scores it honestly, and helps you consolidate it — on your own infrastructure.
Search the way people actually type.
One messy query surfaces every recorded variant of the same thing — because the engine normalises language and units before it matches.
Expands abbreviations: CS → carbon steel, Red → reducer.
Reconciles units: 10 in = 254 mm, 3" = 3 IN.
Strict match — requires both keyword and size, so no irrelevant rows.
| Match | Item code | Description | Group |
|---|---|---|---|
| 96% | MRO-100231 | REDUCER 10 IN CS | dup #1 |
| 94% | MRO-104880 | RED 10" CARBON STEEL | dup #1 |
| 88% | MRO-101907 | REDUCER 254 MM | — |
| 83% | MRO-102544 | REDUCER 10 IN SS | — |
A full health check in about a second.
One click profiles 100% of the records and scores a representative sample — an honest score you can defend, not an inflated one.
Profiles every record (counts, nulls, duplicate codes) — no sampling on the facts.
Representative sample for the rates — so the % matches whatever your team counts themselves.
Score is 6 quality rules; duplicates are reported separately, per commodity.
| Commodity | Records | Sample | Quality | Duplicate % |
|---|---|---|---|---|
| MECHANICAL | 2,889 | 289 | 90% | 29.4% |
| ELECTRICAL | 1,284 | 128 | 88% | 17.2% |
| INSTRUMENTATION | 259 | 26 | 87% | 15.4% |
| DRILLING | 156 | 20 | 91% | 25.0% |
Not just a number — the actual problems.
The headline stays honest and defensible. Alongside it, “What we caught” shows the real offenders your team will recognise instantly.
Biggest duplicate groups — click any card to expand every member (badge count = full group size).
Lowest-quality records — concrete under-specified items, ranked worst-first.
Defensible headline + damning evidence is what wins a proof-of-concept.
Catch the duplicates that hide behind different words.
The heavy artillery for the cases rules and keyword search can't reach — synonyms, reworded items, cross-unit matches — every verdict explained.
Finds “grub screw” = “set screw” and 2 in = 50.8 mm.
A hard-identity guard vetoes false merges — different part numbers stay separate.
Every group shows confidence + matched / differing attributes and a plain-language reason.
| MRO-102310 | GRUB SCREW M6 X 10 |
| MRO-105774 | SET SCREW 6 MM X 10 MM |
| MRO-100944 | BALL VALVE 2 IN |
| MRO-103127 | BALL VALVE 50.8 MM |
Every item, classified to your standard.
One click classifies any row into your ISO-8000-style taxonomy — the noun, the class, and the attributes that actually define the item's identity.
2,818 classes across 9 commodities — noun + modifier resolved automatically.
Extracts identity attributes (the ones that make two items genuinely different).
Reference is upload-and-refresh — no restart to update the taxonomy.
| Item code | Description | Taxonomy class |
|---|---|---|
| MRO-100944 | BALL VALVE 2 IN CS 150# | 🏷 Ball Valve
size: 2 in · material: carbon steel · rating: 150# |
| MRO-100059 | RESTRICTOR 3 IN | |
| MRO-102310 | GRUB SCREW M6 X 10 | 🏷 Set Screw
thread: M6 · length: 10 mm |
Trust your master data again.
Search it, score it, and consolidate it — with AI that never sends a single record off your network.
Search
Every variant of the same item, in one human query.
Assess
An honest, defensible health score in about a second.
Consolidate
Deep AI dedup with explained, reviewable suggestions.
Let's run it on your catalog.