Project summary

Global AI development has a glaring blind spot. Indigenous epistemologies are almost entirely absent from training data. Hemispher-IA, a social innovation initiative based in Panama, is building a practical response to this. The Memoria Viva Ngäbe project develops an offline-first, mesh-networking platform. It adapts Colombia’s Recordari architecture for the Ngäbe-Buglé people across the Panama-Costa Rica border. The system lets communities record, classify, and govern their cultural data. It does this without relying on constant internet or external cloud servers. Over 18 months, the pilot will operate in 10 to 12 communities. It formally employs six young Ngäbe Digital Weavers. The output is not just a local preservation effort. It produces reusable, open-source building blocks to counter digital extractivism in Indigenous territories.

What are this project's goals? How will you achieve them?

The project targets three specific objectives. First, it builds a working technical implementation of the CARE Principles for Indigenous Data Governance. This is not a policy document meant to gather dust. It is a functional dashboard. Through this system, the Ngäbe Council of Elders can technically enforce data classifications as public, restricted, or sacred. Second, the team must prove that collaborative memory systems can function where cell service is practically nonexistent. The Comarca Ngäbe-Buglé and southern Costa Rica have severe connectivity gaps. The solution is a Progressive Web App adapted to sync peer-to-peer over mesh networks. It is optimized for older Android devices with under 500MB of RAM and minimal storage. Third, the goal is to fine-tune a lightweight NLP model for Ngäbere. Roughly 200,000 people speak this language, yet it lacks a digital footprint. The objective is not to build a massive language model. The project will release an open dataset, a basic transcription pipeline, and a methodological playbook. Once this pipeline exists, the marginal cost of adapting it to other under-resourced languages, like K'iche' in Guatemala or Guaraní in Paraguay, drops to near zero. Achievement relies on combining academic rigor with direct advisory from Recordari’s original creators and Universidad del Rosario. Strict ethical auditing via CLET and CAIDP is mandatory. More importantly, the formal employment of local Indigenous talent ensures those who co-design and maintain the system are fairly compensated.

How will this funding be used?

The $150,000 request covers 18 months of working capital. This builds open-source public goods and validates the pilot. The allocation is straightforward. About 68 percent goes directly to human capital. The project rejects the extractive volunteer models common in this space. This budget covers executive direction, a senior technical lead, binational field coordinators, and six Indigenous Digital Weavers paid $450 a month plus tech kits. In the Comarca, a formal contract with benefits is often the difference between a young person staying in their territory or migrating to Panama City for call center work. Roughly 15 percent funds technology development. This includes local servers, PWA development, and the NLP fine-tuning pipeline. Another 7 percent covers field equipment. Specifically, refurbished smartphones, solar power banks, and mesh antennas. The remaining 10 percent handles legal advisory for the binational complexity between Panama and Costa Rica, the CLET ethical audit, and a small operational contingency. Field deployments always encounter unexpected hurdles, and this buffer accounts for that reality.

Who is on your team? What's your track record on similar projects?

Hemispher-IA is a Panamanian-constituted social innovation initiative. It focuses on high-level knowledge brokering and managing scalable portfolios across Latin America and the Caribbean. The team brings together software engineering, Indigenous governance expertise, and academic rigor. Executive direction is led by Wolfgang Friedl. He has over a decade of experience building institutional partnerships for social innovation across Latin America. This includes work with UNICEF, the WHO, and various diplomatic coalitions. Technical and methodological heavy lifting comes from advisors Arthur Picerna and Luisa María Rincón at Universidad del Rosario. They co-created Recordari, an open-source architecture for resilient historical memory in post-conflict Colombia. They are now helping adapt its offline-first data structures for an Indigenous context. On the ground, OLINABP, the Ngäbe-Buglé Future Leaders Organization in Panama, and FAIR Costa Rica serve as the primordial operators. These are Indigenous-led and local responsible-AI organizations with direct legitimacy and a proven track record of community governance in the pilot regions. The governance and ethics board includes CLET for ethical audit and CARE certification, CAIDP for human rights and privacy frameworks, GCRAI for global AI alignment, and the Silicon Valley Certification Hub for auditable AI standards. The track record relies on validated partnerships and deployed frameworks rather than a long list of academic publications. The Recordari architecture actually works in the field. The CLET audit process is established. The AITE principles are proven. Combining all three in a binational Indigenous context is the specific gap this pilot addresses.

What are the most likely causes and outcomes if this project fails?

Integrating Indigenous epistemologies into global AI infrastructure is a multi-decade challenge. It is highly likely the project will not achieve complete continental adoption within the initial 18-month timeline. Legal fragmentation between Panama and Costa Rica alone could cause significant delays. Community adoption might be slower than anticipated. The NLP work might yield a modest dataset rather than a robust model. However, even if the broader vision stalls, the funding will still produce high-leverage, open-source public goods. The CARE implementation code will be available on GitHub for anyone to fork. The offline-first PWA will function in any low-connectivity environment, not just Indigenous ones. The NLP pipeline for Ngäbere will serve as a methodological template for dozens of other under-resourced languages. Finally, the binational replication playbook will openly document the technical, ethical, and social bottlenecks encountered. This will save the next team months of trial and error. The preference is to build durable public goods that survive a partial failure over a flashy demo that disappears when the grant ends.

How much money have you raised in the last 12 months, and from where?

Over the past 12 months, the initiative has secured significant in-kind contributions and seed support. This includes technical advisory hours from the Recordari team, a committed pro-bono ethical audit from CLET, and formal community access plus field infrastructure from OLINABP and FAIR Costa Rica. Preliminary conversations are currently underway with institutional partners like Senacyt in Panama and KAS-PPI regarding co-funding. This Manifund grant would represent the first targeted cash injection specifically for the Memoria Viva Ngäbe pilot. It is necessary to transition from planning and in-kind support to active development, hiring, and field deployment.