Governed AI for Lao institutions,
built for the powerhouse, the river, the survey team and the border

Power purchase agreements: metering, availability and settlement across a border

Laos has built a substantial part of its economy on selling electricity to its neighbours. Generation on the Mekong and its tributaries is contracted to offtakers in Thailand and Vietnam under power purchase agreements measured in decades, with concession terms, tariff structures, capacity and energy payments, availability guarantees and take-or-pay provisions written years before the first unit is delivered. Regional arrangements move Lao power onward through neighbouring systems, which adds parties and adds regulators. Every unit is metered, scheduled and settled across a boundary between two systems whose operators answer to different authorities, keep different calendars and reconcile against their own instrumentation.

What is actually owed is decided by the record: interval metering at the delivery point, declared and achieved availability, dispatch and curtailment instructions, forced and planned outage logs, hydrological conditions and reservoir operation, and the tariff arithmetic applied on top. Where AI assists in forecasting inflows and generation, reconciling metering series, computing settlement, detecting anomalies in a delivery trace or drafting an outage narrative, the platform validates the input before the model runs, validates what comes back against the operator's and the contract's own rules, and writes one immutable ledger row for every call — actor, action, inputs, outputs, model version and verdict. Role-scoped access lets the generator, Électricité du Laos, the foreign offtaker, the lenders and independent engineers each see exactly what their mandate permits. When a settlement is disputed — and on a decades-long cross-border contract it will be — the answer is a record both sides can read rather than two assertions in a meeting.

Mekong monitoring and impact reporting read from outside the country

Dams on the Mekong mainstream and on its tributaries carry consequences downstream: sediment transport, fish passage and fisheries productivity, flow regime and seasonal flood pulse, and water quality. Those consequences are examined by neighbouring states, by the Mekong River Commission under the notification and prior-consultation procedures its members agreed to, by lenders with environmental and social conditions in their agreements, and by researchers and communities with their own instruments. KriftAI takes no position on any particular project or dispute. The operational point is narrower and harder to argue with: the monitoring record is read by parties outside the institution that produced it, and it has to hold up to readers whose interests differ.

Hydrological and sediment series, water quality sampling, fish passage and catch monitoring, environmental flow obligations, resettlement and livelihood commitments and periodic compliance reporting all have to remain traceable to their station, their instrument, their sample and their moment. Where AI assists in gap-filling a series, classifying species data, modelling flow scenarios or drafting an environmental compliance submission, governed capture writes a verifiable record at every step: which data went in, which model version produced the output, what validation it passed, and who accepted it. Assistance from a model is recorded as assistance from a model rather than laundered into a measurement. A series that can be re-run exactly as it stood is a series a downstream government, a basin secretariat and a lender's independent panel can argue about on the merits instead of arguing about provenance.

UXO survey, clearance and the record that says a field is safe

Laos carries one of the heaviest per-capita legacies of unexploded ordnance in the world, concentrated in the eastern provinces but present across much of the country, and clearing it is generational work. The sector runs on records: non-technical survey findings, confirmed hazardous area boundaries, technical survey results, clearance depth and method, quality assurance and quality control, land release and handover certificates, item disposal logs, and accident and victim data. These are not compliance artefacts. A record that wrongly states a plot has been cleared can put a farmer, a child or a construction crew onto ordnance, and the harm is immediate and irreversible.

That is exactly the case for immutable records and mandatory human escalation. Every entry is written once and cannot be silently amended; a correction is a new row that carries its reason and its author, so the history of a claim about a piece of land is never overwritten. Where AI assists — triaging survey reports, reconciling coordinates and legacy bombing data with field findings, prioritising tasks, translating between Lao and English documentation, or flagging inconsistencies between survey and clearance records — the output is validated against the national standards before it reaches anyone, and any output that would affect a hazard classification, a boundary, a clearance depth or a release decision escalates to a named accredited person who makes the call and whose decision is itself recorded. No model releases land. The platform's job is to make sure the human who does has complete, attributable and unaltered evidence in front of them.

External debt, creditor reporting and obligations that outlast the project

Substantial external borrowing tied to infrastructure has left the Lao PDR with debt service and reporting obligations that are, by construction, longer-lived than any single asset. Loan agreements carry covenants, reporting schedules, disbursement and drawdown conditions and security arrangements; concession and project structures carry their own accounts, guarantees and contingent exposures; and creditors, international financial institutions and auditors each need consistent figures reported on their own cycle. Debt sustainability is assessed and reassessed by parties outside the country, and the assessments rest on what has been reported.

The evidence problem is one of consistency across time and across readers. Loan-by-loan registries, disbursement and repayment schedules, currency composition and exchange rate treatment, guarantees and contingent liabilities, and project-level accounts have to reconcile to the same underlying facts whether they appear in a domestic budget document, a creditor return or an international reporting template. Where AI assists in reconciliation, currency translation, drafting reporting submissions or scenario analysis, every call passes the governed chokepoint: input validated, output validated against the ministry's own rules, escalation to a person wherever a figure would enter an external report, and an immutable row recording the whole exchange. Produce the reporting evidence once and let it be read many times, by readers who will still be reading it long after the officials who produced it have moved on.

The Laos-China Railway: transit, customs and freight documentation across borders

The railway running from the Chinese border at Boten south to Vientiane changed how goods move through the country. A landlocked route that was road-bound and slow now carries containerised freight and passengers on a standard-gauge line, with a dry port and logistics facilities outside Vientiane and onward connection toward the Thai network across the Mekong. Every crossing is a documentary event: manifests, declarations, tariff classification and valuation, rules of origin, transit bonds and guarantees, phytosanitary and other permits, weighing and inspection results, and the customs clearance that follows from all of it.

Those documents are prepared by one party, tested by another and settled between customs administrations in different jurisdictions. Where AI assists in classifying goods, extracting data from shipping documents, screening consignments for risk, translating between Lao, Chinese and English paperwork or reconciling a manifest against a declaration, the input is validated before the model runs and the output is validated against the administration's own rules before it reaches an officer, with escalation to a person wherever a determination would change a duty, a release or a hold. Each call writes its own ledger row, so a classification challenged months later can be reconstructed exactly — the document that was read, the model version that read it, and the officer who accepted or overrode the result.

Bank of the Lao PDR supervision, kip reconciliation and digital payments

The Bank of the Lao PDR issues the kip, conducts monetary policy and supervises a comparatively small banking sector alongside microfinance institutions and a payments landscape that has digitised quickly, with QR-based retail payment now ordinary in Vientiane and the provincial centres. Institutions here operate with meaningful foreign-currency exposure, correspondent relationships that carry their own diligence expectations, and cross-border flows tied to power exports, mining, the railway corridor and remittances.

That changes what evidence has to look like. Anti-money-laundering and sanctions screening, customer due diligence, credit and risk models, supervisory returns and multi-currency reconciliation all run through one governed chokepoint, producing evidence in a form the central bank, a correspondent bank's compliance function and an external auditor each accept without a translation exercise afterwards. Screening is enforced rather than advisory: a hit blocks the call, an override escalates to a named person, and the override decision is itself written to the ledger with its reason. A platform does not deliver compliance — the institution carries the obligation — but it makes the obligation tractable by producing better evidence, in Lao and in English, at the moment the decision is made rather than reconstructed afterwards.

On-premise deployment inside Lao facilities

Enforcement, inference and audit logging run inside the institution's own environment — a data centre in Vientiane, a powerhouse control room, a provincial energy or mines office, a bank's own racks, a clearance operator's provincial base. Data does not leave the perimeter in order to be processed, which is the only version of data residency that actually holds.

Air-gapped operation as a supported configuration

For UXO hazard and victim records, supervisory returns, debt registries or contract settlement data, deploy with no outbound path at all. The platform does not phone home for telemetry, licence checks or model routing, so an air gap is a supported configuration rather than a degraded one.

Offline-capable enforcement across mountainous terrain

Governance does not pause when connectivity does. On a survey task in Xieng Khouang, at a remote powerhouse or gauging station, at a mine site in Savannakhet or at the Boten border crossing, validation still executes and the ledger still writes, and records created while disconnected reconcile cleanly when the link returns.

Jurisdictional residency for cross-border power, transit and creditor reporting

Electricity export, railway transit and lender reporting each touch more than one legal regime. Deploy so that data governed by one jurisdiction's rules stays where those rules require, with jurisdiction-specific policy enforced at the same chokepoint rather than in separate systems that drift apart.