A query lands on an officer's desk: does an existing Government Order already cover this exemption, and what did it cite?
Somewhere in the archive, the answer exists. It was issued years ago, by a department that has since reorganised, signed by people who have moved on. It sits inside a scanned PDF, an image of a page rather than text a computer can read. Finding it means remembering it exists, then opening file after file until the right one appears.
Most of the time, someone does find it. But "most of the time" is the problem. The one order that gets missed is the one that contradicts a fresh decision, or the precedent that would have settled a dispute before it escalated.
This is exactly the problem PIPRA built Kuyil to solve. Kuyil reads scanned orders with OCR, extracts their key details, detects the orders they reference, and answers plain-language questions in milliseconds.
The result is that institutional memory stops living in people's heads and filing cabinets, and starts answering questions on demand.
The records themselves are accurate and the drafting is careful. The expertise behind them is real.
The weakness is everything that happens around the document: locating the right one and confirming what it connects to. That work is slow, depends on who happens to be in the room, and does not scale. Every new order added to the archive makes the next search a little harder. And because a Government Order rarely stands alone (it amends or supersedes others, or relies on them), finding one document is only half the task. You also need its chain of references.
When that retrieval step fails quietly, the consequence is real: decisions made without the full picture.
Picture the current workflow. An officer receives a request that hinges on prior orders, and tries to recall which department issued the relevant GO and roughly when. The next move is a shared-drive search by filename, or a question to a colleague who has been there longer. Scanned PDFs get opened one by one, each checked for the right reference. Once a likely order turns up, its "Read" section reveals which earlier orders it depends on, so the hunt begins again for those. If a senior colleague is on leave, the search stalls.
None of this reflects a lack of skill. It reflects a process where human effort is the only index, and human effort does not scale to tens of thousands of documents.
The concrete risks:
What changes is that the archive itself becomes searchable, and answers in plain language.
When a document enters Kuyil, OCR reads the scanned page and turns the image into text. The system then extracts the details that matter, including department, GO number, date, abstract, and full text, and detects the cross-references each order cites, capturing the letter numbers, dates, and departments it depends on.
After that, every order is findable by what it says rather than by who remembers it.
An officer can ask in natural language, such as "health department orders from March 2024" or "the order on a particular exemption", and get the right documents back in milliseconds, with the matching passages highlighted and the references attached as context.
For legal and policy work, PIPRA took this further. Kuyil's legal configuration spans statutes, judgments, and government orders, and uses a two-stage retrieval approach: it first narrows to the documents genuinely relevant to a question, then searches within only those. Answers come backed by specific citations (a section, a GO number, a case reference) rather than vaguely related text. For compliance questions, it can return an actionable checklist.
The goal is one sentence: replace remembering with finding.
The value here is the operational outcome, not the technology itself.
Any officer can retrieve any order in seconds, whether or not they were around when it was issued. Because Kuyil surfaces the references an order cites, a search returns the order and its story: what it amends and what it connects to. Every answer in legal and compliance work points to a specific section, GO number, or case, so it can be trusted and defended.
Knowledge stops retiring with people. The archive becomes a resource the whole department can question, and a new officer can ask the system what previously required years of accumulated familiarity.
Kuyil does not decide anything. It reads and organises so that the people who do decide have the full picture in front of them.
The division of labour is natural. Officers bring judgment and legal interpretation, and an understanding of intent and consequence that no system has. AI brings speed and consistency, plus the patience to read every page and follow every reference without tiring.
In Kuyil's legal workflow this is explicit: the system retrieves and synthesises a cited draft, and final decision authority stays with the officer. The combination beats either alone, because human judgment is applied to complete information instead of partial recall.
Archives only grow. Every year adds more orders, circulars, and amendments, and each addition makes manual search slower while expectations for fast, accurate responses keep rising. RTI requests, audits, and cross-department coordination all assume records can be produced quickly, and increasingly they must be.
The places this pays off:
Better retrieval is the difference between deciding with the full record and deciding with a fragment of it.
A Government Order is one kind of document with a fixed problem: critical knowledge locked inside an unsearchable scan. That problem is not unique to any one department, or to government.
The same framework (read the document, extract what matters, map how documents connect, answer in plain language) retargets to statutes and judgments, to registration records, to any archive where finding the right paper is the bottleneck. Only the data changes.
For the records that run a government, the change is concrete: instead of hunting through boxes and hoping someone remembers, you ask a question and get the right order, with its context, in seconds.
If your team still searches the archive by memory and filename, it is worth seeing what instant, natural-language retrieval would look like on your own records. We are happy to walk you through how Kuyil approaches it.
