Knowledge and semantic engineering means connecting what your organisation already knows and agreeing what its words mean, so people can find an answer without knowing which shelf, folder or person it lives with.
Ask the front desk
In the film, the library is your company’s documents and systems, and its rooms are your teams. The shared dictionary is the meaning your teams agree for their words. Catalogue cards say what each thing is and who wrote it. The red strings are the recorded links between people, things and documents. The front desk is search you can ask in your own words. Theo, the boy waiting on the phone, is a customer or colleague waiting on an answer. When three rooms give one answer, that is the dictionary and the strings working together.
How does the desk look?
Choose one; picking one turns the other off.
Matches the spelling · “money back”
Matched the spelling in one room. The other two rooms use different words for the same thing.
Matches the spelling
Checks the dictionary, cards and strings
People and AI helpers can ask the same desk.
Worth knowing: The desk finds only what has been catalogued, connected and shared with the person asking.
Experts call this: semantic and enterprise searchA simplified illustration, not a measurement.
Watch · 2 min 19 sec
Ten-year-old Theo phones the library: can he get his late fee back? The answer is in three pieces, in three rooms, and every room calls him something different. A short story about connecting what you know and agreeing what words mean.
The idea in plain words
Each one in everyday words first, then the name experts use for it.
Connect what you know. Agree what words mean.
01
Sales says customer, support says client, billing says account holder. Your teams keep their words; the dictionary writes down which ones mean the same, which word means two things, and how kinds of things relate: customers place orders, an order has an invoice.
Experts call this: ontology, taxonomy and business glossary
Worth knowing: A shared dictionary is more than a list of synonyms. It says what kinds of things exist and how they relate, and it keeps each team’s own words tied to one agreed meaning. It is agreed with the teams, and it changes as the business does.
02
Each important document or record says what’s inside, who wrote it, when it was last checked and who looks after it, so people can tell what something is before they open it.
Experts call this: metadata
Worth knowing: Catalogue cards help only while someone keeps them up to date. Cards can be suggested automatically where that is safe, and people check the ones that matter.
03
Record the actual links between particular things, such as this customer placed this order, or this supplier supplies this part. Each link has a verb: wrote, supplies, is about. Then one question can follow the strings instead of five people chasing them.
Experts call this: knowledge graph
Worth knowing: Strings show what people have recorded. Links suggested automatically need a person’s review, and spotting that two records are the same customer, product or supplier is its own careful step.
04
One place to ask that checks the dictionary, the cards and the strings across your apps, and is set up to respect who may see what.
Experts call this: semantic and enterprise search
Worth knowing: The front desk can only find what has been written down, catalogued, connected and shared with the person asking. Asking in your own words works when those words are in the shared dictionary. Search and AI helpers should carry each source’s access rules through, so people see only what they could already open. Checking this is part of the work.
See the difference
Each pin is one particular thing your company keeps a record of. Each red string says how two of them are actually linked: this supplier supplies this part. Ask the board a question and follow the strings it plucks.
The late supplier supplies the gearbox part, which goes into the packing machine, which holds up the open orders.
Worth knowing: The board shows only the links people have recorded.
Experts call this: knowledge graphAgree what words mean
Three teams use three words for the same person. Head office could send a memo, or someone could start a shared dictionary. Try both.
Memo from head office
From today, everyone says customer.
The shared dictionary
customer (also: client, account holder, buyer) = a person or company that buys from us
Preferred in shared reports: customer
The memo says customer. Support still says client, Billing still says account holder, and by lunch the partners team says buyer. Now there are four words, and search misses buyer.
The dictionary also records a word that means two things, so nobody mixes them up.
Worth knowing: Agreeing a preferred word for shared reports is often useful. Ordering everyone to drop their own words tends not to stick, because those words mean something in their work. The dictionary records which words mean the same, and can still mark one as preferred.
Experts call this: shared vocabulary (part of an ontology)Sounds familiar?
If a few of these ring true, the answers exist but nothing links them.
Six plain questions. Answer what you can, and we’ll point to where we’d look first.
Answer any question and the places we’d look first appear here.
Answer “No” or “Not sure” to any question to email the list.
Nothing is stored or sent unless you choose to email it.
What we help with
Eight pieces of work. Start with one area and a few real questions, then add the next.
Agreeing what words mean, with the teams that use them.
A short list of the questions worth answering first, and why each is hard today.
We collect the questions people struggle with, watch where they look today, and list the shelves, words and catalogues you already have.
Experts call this: knowledge and search assessment
One agreed meaning for the words that matter most in that area, each with a named owner.
We sit with the teams that use the words, agree plain meanings, list each team’s own words for them and record how ideas relate. The dictionary is agreed with those teams and changes as the business does.
Experts call this: ontology, taxonomy and business glossary
Also in this area
Saying what each thing is, and keeping that current.
Cards that say what something is, how current it is and who looks after it.
A simple standard for what each document or record should say about itself, and named people who keep the cards up to date. Cards can be suggested automatically where that is safe; people check the ones that matter.
Experts call this: metadata standards, auto-classification
Also in this area
Recording how things are linked.
Linked records, so reports that use the links can count one customer once.
We find where one customer, product or supplier appears under several names and codes, and link them. Links suggested automatically need a person’s review, and spotting that two records are the same customer, product or supplier is its own careful step.
Experts call this: entity resolution, master and reference data
A working answer to that question, with the path of strings that led to it.
We record the links between people, documents, products and places that one important question needs, from your own sources.
Experts call this: knowledge graph design and build
Also in this area
Putting it in front of people and systems.
One search across your apps, using your agreed words.
Search across your apps that uses the shared dictionary, the cards and the strings, and keeps each source’s access rules. It finds what has been catalogued and connected, so we start with the questions people ask most.
Experts call this: semantic and enterprise search, relevance tuning
Agreed codes and mappings for the records that move between systems.
Shared codes and definitions for information that moves between your systems, or to partners and regulators, using sector standards where they exist. Getting systems to read each other’s information the same way takes agreed formats and mappings; the right choice depends on the systems involved.
Experts call this: semantic interoperability, data exchange standards
Answers that point back to your own sources, which people can check.
Giving AI assistants the same dictionary, cards, strings and access rules your people use, so they have your own sources to answer from and point back to, and a way to check their answers. AI helpers can still be wrong, so people check what matters.
Experts call this: knowledge-grounded AI, GraphRAG
Also in this area
For your technical team: we can work with what you already use, for example SharePoint, Confluence, Elasticsearch, OpenSearch, Azure AI Search, Neo4j, Amazon Neptune, GraphDB, dbt and Snowflake, and with open standards such as RDF, OWL, SKOS, SHACL, Dublin Core and schema.org.
Ways to start
A workshop that picks the first questions, and the words that cause the most confusion.
A look at how one shared folder or system is catalogued and searched today.
Agreed words for one team or area, each with an owner.
A small set of strings that answers one real question, with the path shown.
How we work
The same steps as the film: questions first, words agreed with your teams, then cards and strings for what those questions need.
Questions first
We collect the questions people struggle to answer and see where they look today. Those questions decide what gets catalogued and connected first.
Agree
With the teams that use them, we agree plain meanings for the words that matter, keep each team’s own words, and note the words that mean two things.
Catalogue
We write catalogue cards for the documents and records the chosen questions need, and spot the same things under different names.
Connect
We record the strings those questions follow, put a front desk in front of them, try it with the people who asked, and keep the access rules in place.
Keep it tidy
Named people in your teams look after the words, cards and strings, with a simple way to suggest changes. We show them how. Keeping it current is ongoing work, so we start small; we can help with the next area if you want us to.
Where it fits
Illustrative examples, not customer stories: the kind of work this approach suits.
Good to know
Plain answers to what owners and tech leads ask first. Something else on your mind?
Ask us directlyConnecting what your organisation already knows and agreeing what its words mean, so people can find an answer without knowing which shelf, folder or person it lives with.
Search is the front desk. What it can find depends on the dictionary, the catalogue cards and the strings behind it, so that is where most of the work goes.
No. Documents and records can stay where teams keep them. We catalogue them, agree the words and record the links, and the desk looks across them. Some setups copy or convert data, or build a search index of it (a list of what’s where); which one depends on your systems.
Not necessarily at the start, and not for every question. Some questions are answered well by catalogue cards and a shared dictionary in the tools you already have. When questions need to follow many links, a graph database (a database built to store and follow strings) is often the right home for them.
An AI assistant answers from what you give it. With a shared dictionary, catalogue cards and strings, it has your own sources to answer from and point back to. It can still be wrong, so people check what matters.
The desk should show each person only what they could already open. We check that each source’s access rules carry through, for people and for AI helpers.
With a handful of questions your people can’t answer yet, and the words that cause the most confusion. One area first, then the next.
Your own people, named for each area, with a simple way to suggest and approve changes. We show them how, and we can help with the next areas if you want us to.
Heard it before?
Answer: A search box can only find what has been catalogued and connected. Smarter search can guess that client and customer are close, but it can’t know that your Billing team’s account holder is the same person, or that account means a login in IT. That has to be written down.
Answer: That was head office’s memo in the film, and by lunch there was a fourth word. A preferred word for shared reports can help, but ordering teams to drop their own words tends not to stick. Write down which words mean the same instead.
Answer: AI reads what you give it. If your facts are scattered and your words disagree, it guesses. Connected, catalogued facts give it something solid to answer from, and people still check what matters.
Answer: One big pile is still a pile. Finding comes from cards and strings, not from location.
Answer: Start with the questions people ask most and the words that cause the most confusion. Build the dictionary and strings for those first, then grow.
Answer: Knowledge keeps arriving. Cards and strings need simple habits and named owners so they stay current.
Answer: A file in a folder lives in one place. A contract is about a customer, a product, a place and a date at once. Strings let it sit everywhere it belongs.
Tell us a question your people can’t answer yet. We’ll suggest where to start: which words to agree, what to catalogue and which strings to tie.
Or write to contact@algoshred.com
