A Different Use Case per Department
A revenue department and a disaster management authority share nothing in subject matter and one thing underneath: the rules that decide an outcome are written somewhere no software can read.
We already run an AI layer across five live state government portals. See that case study for how the layer behaves on a system in production.
One Rule Before Any of It
Anything the department in the room holds itself. No permission to negotiate, no other agency's cooperation, no waiting. Work that can begin on Monday.
Anything needing another department's data sharing agreement. Valuable, slower, and never the thing a first phase depends on.
Who already holds the data decides what we do first. Most government data projects stall in a memorandum between two agencies. Start with registers the department already owns and a first phase can land inside one financial year.
The Same Three Questions, Everywhere
| Department | What is written down nowhere | What that costs today | First deliverable |
|---|---|---|---|
| Industry | What each scheme really requires, and in what order | Investors wait on officers who know | An eligibility model behind the single window |
| Revenue | Which patterns in the register indicate a leak | The same leak is found again every year | Rules as code, each citing its statute |
| Procurement | What a suspicious award looks like across a body of tenders | Only single files get read | A detection grid over published fields |
| Drafting | Which instruments contradict which | Conflicts surface in litigation | Obligations typed and compared |
| Disaster | Who must act on a given warning, and by when | The map lives with experienced officers | Warnings that produce an owned task list |
| Local bodies | What is pending, and with whom | A diary and a chat thread | Nothing exists without an owner and a date |
| PSU oversight | Which bodies overlap in mandate | Suspected, never evidenced | A roster where every value opens its source |
Scroll the table sideways →
Read the second column down the page. It is the same sentence every time: the rule that decides an outcome was written for a person to apply, and never written in a form software could check. That is the whole of what we do, and it is why the work transfers between departments that otherwise share nothing.
Industry and Investment Promotion
An investor asks what land, what incentive, what clearance, and in what order. The answer exists across a dozen schemes, several agencies and a stack of orders, and lives in the memory of a few officers. Single-window portals often route the paperwork without ever encoding what each scheme actually requires.
The eligibility ontology: every scheme, the conditions it imposes, the document that proves each condition, the officer who signs each stage, and what makes a file complete. On top of that, a single window that can tell an applicant what is missing on day one, and tell an officer which files are decision-ready and which are stuck and why.
Revenue and Arrears
Auditors report large arrears every year. The underlying detail sits in registers that were designed for collection, not for analysis, so the same leak is rediscovered annually rather than prevented.
Public registers normalised into a typed model with provenance on every record, then detection written as plain deterministic rules rather than a model. Every finding cites the statute or rule it rests on and carries its own derivation, step by step, with the figure and the source quote at each step.
Published registers pulled and stored raw first, so every later pass re-runs from the same bytes.
One typed model, provenance per record, and the gaps written down rather than hidden.
Rules as code, no model anywhere in detection, so any finding re-derives live in front of whoever is challenging it.
An assistant that may only quote the sources it was handed, verified word for word, and refuses when it cannot.
Detection must be deterministic in government work. A finding that cannot be reproduced in a meeting is a finding that will be dismissed in that meeting. Rules as code re-derive; a score does not.
Procurement Integrity
Tender data is published and almost never read as a body. Patterns that only appear across thousands of awards stay invisible to the officer looking at one file.
A grid of tests, all of them computable from fields the government already publishes:
- Awards made on a single bid
- Bid windows shorter than the rules require
- Repeated awards that sit just under a delegation limit
- One buyer concentrating its work on one vendor
- The same buyer and vendor pairing again and again
- Corrections that keep pushing a deadline back
- Award values drifting away from the estimate
- Awards clustering at the end of a financial year
- Bidders that share an address
- Set-aside quotas missed against policy
Each hit becomes a finding that cites the exact rule and shows its working.
Legislative and Regulatory Drafting
Rules accumulate for decades. Amendments overlay orders, orders overlay acts, and departments issue clarifications that quietly contradict each other. Nobody holds the whole body in their head, so contradictions are discovered in litigation.
The corpus of instruments modelled as things rather than documents: obligations, thresholds, definitions, exemptions, commencement dates and repeals, each anchored to the clause it comes from. Once obligations are typed, contradiction detection becomes mechanical.
The system's job is to find and show, never to decide. Which rule prevails is a legal judgement belonging to the department. Show both clauses with their sources and the judgement takes an afternoon instead of a courtroom.
Disaster Management and Weather
A warning is not an instruction. A forecast arrives, and the question nobody can answer fast enough is who must do what, where, by when, and under whose authority. That mapping exists, in a standing order and in the experience of officers who have handled the last three events.
An ontology that connects the physical world to the administrative one: gauges, reservoirs, rivers, wards and villages on one side; officers, standing orders, thresholds and escalation ladders on the other. A warning that crosses a threshold then produces a task list instead of a notification. Every line has an owner and a deadline. If nothing happens, the escalation fires by itself.
Which officer, which ward, what action, by when, under which clause of the standing order. If nobody responds, the task escalates automatically.
A gauge reading is tied to the settlements downstream of it, the officers responsible for those settlements, and the threshold in the standing order that makes the reading actionable.
Feeds you already receive, and documents you already have, most of which have never been machine-readable.
This is why a weather system is an ontology problem rather than a data problem. The forecast is already good. What is missing is the written map from a number to a duty.
Local Bodies and Constituency Offices
A ward office or a legislator's office runs on requests: a grievance, a certificate, a works demand, a recommendation. They arrive by every channel, and the record of what happened lives in a diary and a WhatsApp thread. Nobody can say what is pending and with whom.
Case operations where nothing can exist without an owner and a next date. Every request typed, routed and aged, with the standing rule that silence escalates rather than sleeps. For a constituency office, that becomes an honest picture of what was asked, what was done, and what is stuck, which is also the only defensible answer at the end of a term.
Public Sector Oversight
A state may hold hundreds of public sector undertakings, boards, corporations, societies and authorities. Their mandates were written across decades. Overlapping objects clauses, dormant bodies and duplicated functions are widely suspected and rarely evidenced.
A verified roster where every value opens the document it came from, and nothing is invented to fill a gap. Where a value cannot be sourced it stays empty and the gap is reported, because an empty cell is itself a finding. Overlap is then shown as evidence, with the objects clauses of two bodies set side by side verbatim, never as an assertion that they duplicate one another. Bodies mandated once per district or per university are marked as structurally required so they are never mistaken for duplication.