ChatLima is the conversation inside Lima Juno: the part of Juno that answers questions.
31 open purchase order lines worth $348,000 sit inside their cancellation window at plants 1210 and 1220, and the window closes on Friday 18 September.
A reader who wants to trust a figure finds its evidence on the same screen as the figure. Product.
The ontology terms matched in the sentence, so the meaning can be checked at a glance.
Each system that took part, named on the answer: the ERP, the second ERP, the planning suite.
One click opens the query. An analyst reads it, copies it and runs it in their own tool.
Seconds from the question to the answer, so a slow source shows up early.
The plants, categories and columns the asking person is cleared for, stated on the answer.
How sure ChatLima is of its reading. Where a word allows two readings it asks one question; where the evidence is thin it says so.
Two steps settle the meaning; three steps fetch the facts.
A shared, confirmed meaning is what turns a plausible answer into a correct one.
Built-in ontology. Your columns are mapped onto it; your team confirms.
ProductPossible lift in generative AI accuracy from attention to semantics, by 2027.
Gartner · 17 June 2025Correct answers from raw tables vs a knowledge graph of the same data.
data.world · 13 Nov 2023Every query runs with the asker’s rights, and the answer says what applied.
| The person | The rows they may read | The columns their roles allow | The answer to one question about a supplier |
|---|---|---|---|
| D. Okafor · buyer | Fasteners and metals at plants 1210 and 1220 | Every purchasing column, the contract unit price among them | 486 lines due · 36 late · 7.4% · $1.24M |
| J. Ridley · supply chain director | Every plant and every category | Delivery and value columns; the price column sits with purchasing | 1,412 lines due · 104 late · 7.4% · $3.86M |
| Delta Fasteners Ltd · supplier partner | The order lines that belong to this supplier | Quantities, dates and their own performance | 1,412 lines due · 104 late · 7.4% |
One definition of a late line, three scopes; each figure is correct for the person who asked. Showcase estate.
The rules are built into the query itself, so a question that reaches three systems carries one scope into each of them. When J. Ridley asks for the contract unit price, the answer names the column, names the purchasing role that holds it, and gives the rest of the table whole. He knows what exists and whom to ask.
Three more questions on the same thread, each one shorter than the last.
Delta Fasteners Ltd holds 11. The other 12 are at plant 1220.
filter: plant 1210context keptAll with Delta Fasteners Ltd, confirmed after 5 October.
column addedin your scope“Monday cancellation review”: three questions, re-run any time.
walkthroughshareableThe same conversation carries comparisons, causes and forecasts, each with its rows on show.
The same measure against last week, last quarter or the same week last year, from the history your sources hold.
Candidate causes ranked by how much of a change each accounts for, with the rows behind each and a confidence beside it.
The next weeks for a measure you follow, given as a range, re-run each night and kept beside what happened.
Below the confidence floor your team sets, ChatLima says the evidence is thin. Causes, trends and forecasts in detail
Five places to ask, one meaning underneath, one record of who asked what.
Each surface applies the same mapping, the same vocabulary and the same access rules, and each question leaves the same audit line, so a planner asking in Teams and an analyst asking in the ChatLima screen receive one figure for one definition.
You bring the questions and the answers you expect; we bring the platform, the method and the score.
Any source registered through the data fabric: SAP, Oracle, planning suites, warehouse and logistics systems, lakes, cloud warehouses, databases and files. One question can read several and join the results.
Both read the same ontology. The purchase order line L. Tan asked about on the Wednesday, when the answer was that week’s 31 lines, is the object the PO cancellation agent acted on the following Monday, when a customer push-out had left 38 lines above the need, under a policy written in the same words. Chapter 6 shows the shared map.
A person marks it and corrects the term it misread. The correction is stored, the verified question-and-answer pair joins the set, and the next answer uses it. The feedback loop
Two weeks, two of your systems, fifty questions with the answers you expect. We score accuracy, speed and access correctness, and you keep the scored report, the question set and the query behind every answer.
