A realistic five-year agenda

The list

1. A national personal-data-protection law with a real regulator. The base layer on which every other piece sits. Without this, the country is operating without speed limits or seatbelts. Pair the law with an institution that has audit power and budget.

2. A standing inter-platform working group on remittance data. Convened by the central bank, attended by the major remittance platforms, with the explicit task of designing a privacy-preserving aggregation layer that unlocks economic nowcasting and fairer credit. Eighteen months from convening to first dashboard.

3. A public Nepali speech corpus. Multi-dialect, multi-gender, conversational, with metadata. Funded by an NGO–university partnership, hosted on a public platform, kept maintained for at least five years. This single asset unlocks half the Nepali voice AI applications worth building.

4. A working Devanagari OCR programme for the Supreme Court archives. Five decades of judgment, made searchable. The single most-leveraged public-good OCR project in the country. Treatable as a national heritage project as much as a tech one.

5. A country-complete flood early-warning system with last-mile dissemination. Take DHM’s existing capability, extend coverage to every basin that matters, and invest as much in dissemination (SMS, FM, ward-level WhatsApp groups) as in modelling. Treat the SMS gateway and the volunteer network as core infrastructure, not afterthoughts.

6. A national agricultural data hub. Maintained by NARC with academic partners. Labelled image sets for the top twelve crops, district-level price and acreage time series, satellite-derived vegetation indices, weather. Free for use by Nepali researchers and product teams.

7. A national health data exchange with consent and audit infrastructure. This is the largest single piece of public-good AI infrastructure available to build. Three to five years, requires political will and careful design, and unlocks decades of subsequent work.

8. A clear AI-in-decision-making policy for government. Establishes what AI may be used for in administrative decisions, what citizens are entitled to know, contest, and appeal. Without this, every ministry adopts AI separately and badly.

9. Modest national AI compute, hosted for public-good workloads. Tens of GPUs, not thousands. Hosts the citizen-data workloads, the national health AI inference, the disaster-response inference. Available at no cost to Nepali researchers as a public good.

10. A patient-capital seed fund for Nepali AI companies focused on local problems. Small by global standards — even US$10–20 million would matter — with deal terms and timelines designed for product companies rather than software services. Government and donor anchored, with private participation as it matures.

11. A working AI procurement framework for the public sector. Defines disclosure, evaluation, vendor accountability, and a preference for credible Nepali firms on public-good work. A specific small-firm path that does not require the same procurement weight as a major IT contract.

12. A standing AI-and-society dialogue. Not a one-time committee. A real public forum — convened by an academic institution, attended by ministry officials, founders, civil society, and the press — that meets at least quarterly and has visible influence on which questions get asked first.

What the country looks like if this works

If most of these come together by 2031, this is the shape of the country:

• A Nepali in any district can speak in Nepali to their phone and have a working transcription, with a model that runs on the device and was fine-tuned on a public Nepali corpus.
• The Supreme Court’s archives are searchable in seconds. So are the country’s land records. Court delays drop because clerks can find cases.
• The cooperative in Surkhet uses an AI-assisted credit recommendation, calibrated on Nepali data, to lend to members the formal banking system never reached.
• A flood warning issued by DHM reaches every mobile in the affected basin inside five minutes, in the affected dialect.
• The health worker on the trail to Jumla dictates her visit notes in Nepali, gets them transcribed on the phone, and the data syncs when she reaches a tower.
• The Nepali AI ecosystem has perhaps thirty product companies that each employ ten to fifty people, doing real work on Nepali problems, funded by a mix of domestic and regional capital.
• The diaspora is involved enough that the country’s AI direction is set partly by people who have built at scale, but the centre of gravity is in Kathmandu and the provinces, not abroad.

This is not a utopia. It is a country where the technology is being used to solve problems that matter here, by people who understand here, under rules the country set deliberately.

What you can do, this week

If you have read this far, you are part of the small set of people in Nepal who has thought seriously about all of this. The first ask is small: pick the one thing on the agenda above that maps to your current role, your training, or your network. Make a single small move toward it this week — a conversation, a draft, a meeting request, a github commit, an article. The agenda is only as real as the people who carry it forward.

Closing

This is the last section of the course. Thank you for reading.

If the course has done its work, you should leave it with three things: a clearer sense of where AI actually meets Nepal today; an opinion about where it should go; and a small but durable sense that the questions are open — that the future of AI in the country is not yet decided, and that the decision is partly yours to make.

That is more than enough to begin.