Using AI to Build Trust in Civic Education
MEWA is an AI-powered assistant built for Latarnik Wyborczy, one of Poland’s leading civic education platforms. Using a Retrieval-Augmented Generation (RAG) system, the chatbot delivers transparent, verifiable electoral information to help citizens make informed decisions and strengthen trust in public knowledge.
Latarnik Wyborczy, one of Poland’s most recognized civic education platforms, wanted to improve access to reliable electoral knowledge — especially among younger voters. But with public trust in AI at an all-time low, building a chatbot that could deliver facts with transparency and neutrality was a complex challenge.
The team at the Center for Civic Education (CEO) set out to create a chatbot that could answer common questions about elections and explain party stances — all using credible information, no editorial bias. The assistant had to support ethical data use and operate under a strict cost-per-query ceiling, with speed, trustworthiness, and scalability on top.
What sets Nomtek apart is their cross-functional approach: they assembled a diverse team with expertise spanning UX, development, and product strategy to ensure that the solution would be practical, scalable, and aligned with our long-term goals. Their collaborative mindset and ability to explain complex topics in an accessible way made the entire process feel like a true partnership.
We’re excited to continue this journey together and can confidently recommend Nomtek as a reliable, proactive, and thoughtful partner for any AI-driven product development.
Monika Prus-Głaszczka, Head of the Civic Engagement Department at Center for Citizenship Education
“We had a tight timeline and high expectations — Nomtek delivered. The app became a key part of our sales pitch and helped us open important conversations with partners and investors. What impressed me most was how quickly they understood our goals and adapted the design to match exactly what we needed for live demos.”
Mark Reimer, Co-Founder at Hamelin Tech
Scope of Work
The project combined civic expertise with technical precision. We translated educational goals into a reliable AI system architecture that supports transparent data use and smooth interaction. Every design and engineering decision aimed to create trust — from how information is sourced to how it’s presented.
User and Data Discovery
Workshops with CEO defined user needs, tone, and the structure of trusted data sources. The process established the foundations for factual consistency and a tone aligned with civic education standards.
Backend and RAG Pipeline Development
A robust backend was built in Python with LlamaIndex and vector databases to enable fast, accurate data retrieval. The RAG pipeline handles data loading, embedding, and indexing for factual grounding in every generated response.
Document Processing and Prioritization
Data from PDFs, CSVs, and text files was converted into vectorized form and ranked by credibility and freshness. This hierarchy ensures the assistant references only verified, current information.
Query Routing and Classification
A multi-step query system classifies inputs, filters sensitive or opinion-based content, and routes questions to the correct data sources — reducing hallucinations and improving response clarity.
UX and Chat Interface Design
The chat interface was designed to reduce cognitive load and increase transparency. Users see clear disclaimers, source citations, and concise fact-based responses.
Evaluation and Quality Assurance
We implemented a continuous testing framework using RAGAS and custom evaluation sets to measure faithfulness, recall, and bias. Promptfoo and regression tests ensured consistent performance across updates.
AI Red Teaming and Security Testing
Following the OWASP AI Testing Guide, we conducted targeted AI red teaming to identify vulnerabilities like prompt injection or manipulation, strengthening system security and trustworthiness.
Cost Optimization and Deployment Readiness
Performance monitoring and analytics tools were integrated to track model behavior and query cost, supporting long-term scalability and cost efficiency.
Scope of Work
The project combined civic expertise with technical precision. We translated educational goals into a reliable AI system architecture that supports transparent data use and smooth interaction. Every design and engineering decision aimed to create trust — from how information is sourced to how it’s presented.
User and Data Discovery
Workshops with CEO defined user needs, tone, and the structure of trusted data sources. The process established the foundations for factual consistency and a tone aligned with civic education standards.
Backend and RAG Pipeline Development
A robust backend was built in Python with LlamaIndex and vector databases to enable fast, accurate data retrieval. The RAG pipeline handles data loading, embedding, and indexing for factual grounding in every generated response.
Document Processing and Prioritization
Data from PDFs, CSVs, and text files was converted into vectorized form and ranked by credibility and freshness. This hierarchy ensures the assistant references only verified, current information.
Query Routing and Classification
A multi-step query system classifies inputs, filters sensitive or opinion-based content, and routes questions to the correct data sources — reducing hallucinations and improving response clarity.
UX and Chat Interface Design
The chat interface was designed to reduce cognitive load and increase transparency. Users see clear disclaimers, source citations, and concise fact-based responses.
Evaluation and Quality Assurance
We implemented a continuous testing framework using RAGAS and custom evaluation sets to measure faithfulness, recall, and bias. Promptfoo and regression tests ensured consistent performance across updates.
Solution
Together with CEO, we built MEWA — an AI-powered chatbot that delivers clear, verifiable answers about voting processes and political programs. The system uses a retrieval-augmented generation (RAG) architecture to combine verified data sources with large language models, grounding every response in factual, up-to-date information.
MEWA will be integrated into the Latarnik Wyborczy platform, designed for transparency and ease of use. It’s currently in pilot testing and on track for full release in 2027 — supporting citizens with factual, accessible electoral information at scale.
MEWA sets a new benchmark for responsible AI in civic tech — reducing misinformation and promoting transparency in electoral education. Built for long-term maintainability and scalability, the system helps CEO engage younger audiences and foster informed participation in democratic processes.
What sets Nomtek apart is their cross-functional approach: they assembled a diverse team with expertise spanning UX, development, and product strategy to ensure that the solution would be practical, scalable, and aligned with our long-term goals. Their collaborative mindset and ability to explain complex topics in an accessible way made the entire process feel like a true partnership.
We’re excited to continue this journey together and can confidently recommend Nomtek as a reliable, proactive, and thoughtful partner for any AI-driven product development.
Monika Prus-Głaszczka, Head of the Civic Engagement Department at Center for Citizenship Education
What sets Nomtek apart is their cross-functional approach: they assembled a diverse team with expertise spanning UX, development, and product strategy to ensure that the solution would be practical, scalable, and aligned with our long-term goals. Their collaborative mindset and ability to explain complex topics in an accessible way made the entire process feel like a true partnership.
We’re excited to continue this journey together and can confidently recommend Nomtek as a reliable, proactive, and thoughtful partner for any AI-driven product development.
Monika Prus-Głaszczka, Head of the Civic Engagement Department at Center for Citizenship Education
Team Composition
A focused team of backend developers, AI specialists, and UX designers collaborated closely with CEO’s civic education experts. This agile setup enabled rapid iteration, accurate handling of complex queries, and alignment with strict ethical and technical requirements.
AI Engineers
UX/UI Designer
QA Engineer
Product Manager
Frontend Developer
