Ministries of finance · DMOs · IFIs
Debt sustainability analysis
Independently reproducible DSA with a published methodology and a validation gate against official results.
Built · liveGovTech · analytics, AI and implementation
Public-sector analysis that survives a hostile reviewer.
We build analytical systems for governments, multilaterals and public institutions — the kind that get quoted in a parliamentary committee, challenged by a creditor, or audited two years after the people who built them have moved on.
That setting rules out most of how AI is currently sold to the public sector. A model that cannot show its working is not defensible in front of a public accounts committee, and a system that is not defensible is not buyable at any price. So we build to four rules, and we hold to them even when it makes the product slower to ship.
- No language model ever produces a number that reaches a user. Figures come from a deterministic engine or a human-reviewed observation. Enforced structurally: our import graph is tested, and a build fails if a computation module can reach a model SDK.
- Nothing publishes until the validation gate passes. Where an official published result exists, our engine must reproduce it within tolerance before it is allowed to serve output at all.
- A placeholder is never rendered as though it were real. A missing value renders as missing and says it is pending. An unmeasurable target is NOT_EVALUABLE, never a pass and never a breach.
- Provenance is not a feature. Every number carries its source, vintage, extraction method and reviewer. If you cannot trace a number, we do not display it.
The proof, not a case study
Open DSA
An independent, open-source reimplementation of the IMF's Sovereign Risk and Debt Sustainability Framework, built with the Institute of Political Economy and piloted on Sri Lanka.
Debt sustainability analyses drive austerity decisions affecting millions of people, and almost nobody outside the IMF can reproduce them. Open DSA reproduces the published arithmetic, then lets anyone vary the assumptions and see exactly what changes — with every figure traceable to the table it came from.
What it currently says about itself
We transcribed the IMF's combined fifth and sixth review of Sri Lanka's EFF programme and asked our engine to reproduce its published debt path. It does not — and the tool says so, in public, on its own front page.
The divergence reaches 15.8 percentage points of GDP by 2035, and it is fully accounted for: the published table does not contain the projected exchange rate path that its own decomposition depends on. The IMF attributes 12.6pp of cumulative debt reduction to a revaluation term whose driver it never publishes. The published analysis does not contain enough information to reproduce itself.
Validation gate: NOT VALIDATED
Publishable: no
Open the workbench →
Methodology & validation report →
We are showing you a red gate on purpose. Any vendor can demonstrate a green dashboard. The question worth asking is what their system does when it cannot substantiate a number — and whether you would ever find out.
What we build
Instruments, not dashboards
Each of these is a system we can build to the four rules above. One is already built and linked; the rest are scoped engagements, and we will tell you plainly which parts are analysis, which are engineering, and which are a multi-week data-assembly task that no tool can shortcut.
Sovereign finance and debt
DMOs · creditor committees · advisors
Restructuring negotiation simulator
Creditor-by-creditor NPV haircuts, comparability-of-treatment tests, scenario-weighted recovery.
Ministries of finance · SAIs
SOE fiscal risk monitor
State-owned enterprise contingent liabilities — the exposures that actually break sovereign balance sheets.
Treasuries
Cash & liquidity forecasting
Treasury Single Account forecasting and borrowing-calendar planning to reduce idle balances.
Ministries of finance · climate funds
Climate contingent liability modelling
Disaster-loss calibration feeding contingent credit and catastrophe-instrument analysis.
Ministries of finance · parliaments
Fiscal rule compliance monitoring
Continuous, automated evaluation against national fiscal rules, with every breach traceable.
Revenue and integrity
SAIs · procurement authorities
Procurement anomaly detection
Single-bidder patterns, bid-rigging signatures and cross-contract price benchmarking, with every flag explainable to a supplier.
Customs · revenue authorities
Trade mirror analytics
Declared imports compared against partner-country reported exports, using public trade data, to locate revenue leakage.
Revenue authorities
Tax gap estimation
Sectoral gap decomposition and compliance targeting, with the estimation method published.
Registries · anti-corruption bodies
Beneficial ownership analysis
Shell-company and conflict-of-interest detection across company and procurement registries.
Donors · finance ministries
Aid flow reconciliation
Donor commitments against disbursements against budget execution, reconciled and attributable.
Social protection agencies
Registry deduplication & targeting
Inclusion and exclusion error measurement for subsidy and transfer programmes.
Statistics, service delivery and AI systems
Central banks
Reserve adequacy & FX analytics
Reserve adequacy metrics and intervention-efficacy analysis on a reproducible basis.
National statistics offices
Nowcasting from alternative data
Faster indicators from satellite, shipping and mobility data, published with their error bands.
All segments
Document intelligence over official corpora
Extraction across Article IV reports, budget documents, audit reports and procurement notices — with human sign-off before any figure is published. Consistent with rule 1.
Parliaments · budget offices
Budget question answering
Natural-language access to budget and spending data, where every answer cites the line it came from.
Line ministries · municipalities
Multilingual citizen service agents
Service agents with defined escalation paths and a complete audit trail of every interaction.
Ministries · municipalities
Budget transparency portals
Citizen-facing publication of budget execution, held to the same provenance discipline as everything above.
Who we build for
And what each institution is actually judged on
We scope against the objective the institution is measured against, not against a technology category.
| Institution | What the work has to move |
|---|---|
| International financial institutions and MDBs | Analytical rigour, reproducibility, methodological transparency under external challenge |
| United Nations agencies | Cost per beneficiary, aid effectiveness, fraud reduction, SDG reporting |
| Ministries of finance and debt management offices | Revenue mobilisation, debt service cost, programme compliance, avoided default |
| Central banks | Reserve adequacy, forecast accuracy, intervention analytics |
| Supreme audit institutions | Recovered funds, procurement savings, findings that hold up on appeal |
| Line ministries | Service coverage, waiting times, stockouts, workforce planning |
| National statistics offices | Survey processing cost, timeliness, data quality |
| Municipal and city government | Permitting throughput, backlog, citizen satisfaction |
| Sector regulators | Market surveillance, tariff modelling, enforcement evidence |
| Debt advisors and creditor committees | NPV haircut analysis, comparability of treatment |
How we work
The uncomfortable parts, stated up front
We will tell you what we cannot do
Some inputs are not automatable. Assembling an instrument-level debt service schedule from restructuring memoranda is a multi-week analyst task, and we say so rather than shipping a plausible-looking placeholder.
We publish the methodology
The method is written down, versioned, and served from the same repository as the code. If the published page and the implementation could disagree, one is wrong and nobody would know which.
Deviations are documented, not tuned away
A documented 0.8pp deviation from an official result is a stronger position than an unexplained 0.1pp agreement. We do not adjust assumptions until the output matches.
You own it
Open source where it should be. Deployable in your own environment. No dependency on us to reproduce a number you have to defend.
TAI Labs has worked with United Nations agencies and with government and government-affiliated bodies across the United States, Europe and Asia. Specific engagements are covered by client confidentiality; we can discuss relevant references directly under NDA.
Request a demo
Tell us what you need to be able to defend
The most useful first conversation is usually about a specific number that somebody is going to be asked a hard question about.