May 21
Impact Lab: Odporna Demokracja Info session
Join
Fifteen organizations across 10 countries, working on proven interventions in democracy, healthcare, education, climate, agriculture, freedom of expression, and accessibility. All of them with a working solution and a specific technical problem that AI can solve. This is the first ever cohort of the AI Impact Scaling Program, and this is who made it in, out of 80 candidates.
We spent months getting here: scouting organizations that had not applied yet, reviewing applications, conducting structured interviews across time zones, sometimes five in a day, with teams dialing in from Lagos at midnight or Lima at 6am. We argued about organizations that were hard to rank because they were genuinely impressive. We changed our minds more than once.
What genuinely surprised us was how prepared the organizations were. Not polished, but honest: clear about what their prototypes could and could not do, specific about the data they were working with, and unusually self-aware about what scaling their intervention would actually require. That level of clarity is rarer than you would expect from an application process.
We built this cohort on what we learned from the 31 organizations in our pilot program since December 2025.
Watch the story of one of our program participants, Brazil Flying Labs:
Over 12 months, each organization gets three things working in parallel.
First, a dedicated tech partner. Our team spends weeks, sometimes months, on each match. We are looking for a team with the exact expertise the project needs.
Second, specialized infrastructure: dedicated hubs with volunteer AI experts in data readiness, cybersecurity, and responsible AI, with one-on-one consulting available on request. This matters because the problems that derail AI projects in the nonprofit sector are rarely the ones that appear in a proposal.
Third, a peer community. Fifteen organizations, most solving completely different problems in completely different contexts, all asking the same question: how do you take something that works and make it work for ten times as many people?
Together, the 15 organizations in this cohort already reach more than 22 million people across their proven interventions — from 13 million people with access to safe water guidance in India, to 4.6 million citizens using electoral tools in Poland, to 1.4 million people safer from housing collapse across 6 countries. One platform alone, Referandom, reaches 10 million people every month. These are working solutions, already in the world, already producing results. The program exists to help them reach even more.
Poland's largest educational NGO, running Electoral Lighthouse — a voting advice platform reaching 4.6 million citizens per election cycle. They are integrating MEWA, a RAG-based electoral chatbot built by Nomtek, to give voters real-time source-cited answers to political and procedural questions. Target deployment: Poland's 2027 parliamentary elections.
8,000 smallholder farming families reached, with an independently validated 73% net income increase, 58% reduction in crop loss, and 70% savings in water and fertilizer in year one. Their WhatsApp AI advisory system will decouple agronomic guidance from human bandwidth. Kheyti's target is 1 million farmers by 2033.
A cooperative connecting 75+ independent newsrooms and 50,000+ individuals across Europe, translating journalism into 20 languages while keeping traffic with the original publisher. They are replacing commercial LLMs with self-hosted sovereign AI infrastructure and building a multilingual semantic recommendation engine.
Builders of Referandom, a civic platform reaching 10 million people annually by turning parliamentary records into accessible content. CDR is scaling Policyful, its open-source policy intelligence platform, into Türkiye, Bulgaria, Poland, and Brazil with a grant already secured.
60,000 Water Quality Champions trained since 1994, reaching 13 million people across 8 Indian states. Designated National Thematic Lead for Water Quality by the Government of India. JalXChange, their live WhatsApp RAG bot built with IBM, already has 44,000 students engaged in Assam alone. The program expands geographic and multilingual coverage toward a digital public good architecture.
$1.6 million in surplus medical supplies redirected to 25+ safety-net clinics serving 180,000+ patients since July 2024, with a 90%+ delivery acceptance rate. They are building ReadyMedAI: an agentic matching engine routing surplus based on equity-weighted need, plus an open-source toolkit for other health systems to replicate the model independently.
CerviScanner screens for cervical cancer on a standard Android phone with no specialist and no lab infrastructure. A Rwanda pilot screened 260 women with 84.5% diagnostic accuracy and 92.2% specificity. Multi-site expansion across Rwanda and Tanzania is approved for 2026. The program builds the Android-native version, offline capability, and a synthetic data pipeline for broader population coverage.
1.4 million people are safer from housing collapse over 20 years across 6 countries, with 283,000+ households served. They are evolving Tara, their live RAG chatbot in the Philippines, into an agentic system that generates personalized construction guidance based on each household's specific risk profile.
Every intervention selected by one metric: tonnes of phosphorus reduced per euro spent. Eight programs running across Sweden and Poland since 2013. BalticEye AI will direct enforcement resources to the highest-impact agricultural entities across all 16 Polish regional inspectorates. Pilot with Poland's national inspectorate confirmed for 2026.
A regional nonprofit defending freedom of expression across the Americas, with Liberties360 actively monitoring threats to press freedom in Ecuador and reaching approximately 1 million people indirectly. The program automates manual data collection and classification and expands the system to 5 to 8 Latin American countries.
1,500+ guided journeys completed for 400+ visually impaired people across 22 London boroughs, covering the first and last mile that Freedom Pass and Dial-a-Ride do not. VIPA, their AI voice assistant, already handles bookings entirely by voice. The program prepares it for national rollout.
1.7 million trees monitored across 5,000 acres in 6 Indian states, with detection accuracy above 85% for small trees and 92% for mature trees across 14 partner organizations. The program automates the full processing pipeline and builds dashboarding for carbon financiers and partner organizations to access outputs directly.
Neurodiversity Pride Day active in 58 countries, reaching 2.1 million people in 2025 alone. Their volunteer base of 180+ people is 85% neurodivergent. NeuroScale will automate volunteer onboarding, screening, and skill-matching across 8 roles and 20+ global projects — infrastructure that other volunteer-driven organizations could replicate.
Tabshoura, their free digital learning platform, has reached 68,000+ learners across 230 institutions and is now the de facto emergency education system for 1.2 million displaced Lebanese. The program builds an AI layer from scratch: a learning path recommendation engine, multilingual chatbot support in Arabic, French, and English, and an offline-first mobile rebuild for low-end devices.
MiWawa combines maternal mental health screening with early childhood development support across two WHO-validated interventions, piloted with 3,000 individuals in Lima. The program builds AI models to detect risk trajectories and prioritize community health worker follow-up before crisis point, not after.
Specialized hubs, workshops, and the peer community are open now. It will be followed by matching with tech partners.
We will document what we learn from this cohort the same way we documented the pilot: honestly, with the things that did not work alongside the things that did.
If your organization has a proven intervention and a specific technical problem that AI could solve, apply at techtotherescue.org.
