A Polish academic-government consortium built PLLuM, a homegrown LLM, piloted in Częstochowa from August 2025 and rolled into the national mObywatel app by December 2025. Press testing found it gave an outdated answer identifying Poland's own Prosecutor General.
Additional Ministry of Digital Affairs allocation for public-administration rollout (following Częstochowa pilot)
~200,000 residents
Częstochowa pilot city population (August 2025 pilot agreement)
Details
Maturity
Scaling
Promoter
Ministry of Digital Affairs (Ministerstwo Cyfryzacji) with NASK and the PLLuM academic consortium
Period
Consortium project since 2023; Częstochowa pilot signed August 2025; national mObywatel rollout from 31 December 2025
Keywords
e-government, large language models, digital identity, public administration
Context
Six (later eight) Polish public research institutions — including Wrocław University of Technology, two Polish Academy of Sciences institutes, NASK, the Information Processing Institute and the University of Łódź — built PLLuM, a homegrown Polish-language model (8–70B parameters, using mixture-of-experts and retrieval-augmented generation), funded with over 14.5 million zł and explicitly aimed at reducing public administration's reliance on foreign AI vendors.
Objectives
The project sought a sovereign alternative to foreign LLMs for Polish public administration, first for drafting official letters, triaging resident inquiries and summarising documents, and later as a citizen-facing assistant inside the national digital-identity app.
Activities
Częstochowa (population approximately 200,000) signed an agreement in early August 2025 to become the first Polish city piloting PLLuM. The Ministry of Digital Affairs then allocated a further 19 million zł (about €4.5 million) to extend PLLuM into public administration broadly, and from 31 December 2025 a PLLuM-based assistant went live inside the national mObywatel digital-identity app — described by Deputy Prime Minister Krzysztof Gawkowski as the first national-model government chatbot in Europe.
Results
Independent Polish outlets that tested the live assistant found real errors: asked who Poland's Prosecutor General is, PLLuM answered 'Dariusz Barski', who had been removed from that post in January 2024, and only correctly named the actual officeholder, Dariusz Korneluk, when the question was phrased differently — an inconsistency that ChatGPT did not make in the same side-by-side test.
Conclusions
No formal accuracy or citizen-satisfaction results from the Częstochowa pilot or the national rollout have been published as of mid-2026.
Implementation
Indicative cost
High (€500k–€5M) — Over 14.5 million zł in consortium project funding plus a further 19 million zł (≈€4.5 million) Ministry of Digital Affairs allocation for public-administration rollout.
Time to results
Medium (1–3 years) — Consortium project underway since 2023; Częstochowa pilot signed August 2025; national mObywatel rollout live from 31 December 2025.
Staffing & skills
Multi-institution public consortium: Wrocław University of Technology, two Polish Academy of Sciences institutes, NASK, the Information Processing Institute, University of Łódź, later joined by two more institutions, Ministry of Digital Affairs (Ministerstwo Cyfryzacji) providing oversight and further funding
Conditions for success
City-level pilot (Częstochowa) before national rollout, a documented stepwise replication path
Sovereign-model design (8–70B parameters, MoE, RAG) explicitly aimed at reducing reliance on foreign AI vendors
Substantial dedicated public funding (14.5M zł consortium funding plus 19M zł further Ministry allocation)
Common failure modes
Independent journalistic testing found a factual accuracy error (outdated Prosecutor General) that a comparison model (ChatGPT) did not make in the same test
No formal accuracy or citizen-satisfaction metrics have been published from either the pilot or the national rollout
Where it fits
Governance type
national ministry with academic consortium
Scale
national (piloted at city level first)
Income level
high-income (Poland)
Data sources
Where this practice's information was retrieved from, and when.
Bogotá's WhatsApp-based AI assistant Chatico logged 3.6M+ conversations since 2023, handled 1.6M queries during a water-rationing crisis, and carried 147,000+ …
Singapore's OneService chatbot lets residents report municipal issues like potholes or broken streetlights via WhatsApp/Telegram in plain language. An NLP …
