StepHow tests citation-first AI assistant with Tallinn in govtech push
For many city officials, the hardest part of answering a public query is not the answer itself. It is finding the correct document, checking whether it is still valid, and proving where the answer came from.
StepHow, a B2B AI company led by CEO Jasper Hwang, is now trying to solve that problem in one of the world’s most digitised city governments. The company has been selected for the City of Tallinn’s Test in Tallinn innovation programme and has signed a memorandum of understanding with the Estonian capital to begin a proof of concept for its AI-based data search and chatbot product, Wissly.
Also Read: How to drive business innovation with AI-powered data analytics
The project will test whether Wissly can help Tallinn officials search across multiple administrative systems, manage documents more efficiently, and generate evidence-backed answers from public records. Rather than simply producing a conversational response, the system is designed to show where each claim comes from, with direct links to source documents.
That detail matters. As governments experiment with generative AI, the central question is no longer whether chatbots can produce polished answers. It is whether public agencies can trust them when accuracy, auditability, and citizen rights are at stake.
A test inside a mature digital government
Tallinn is a meaningful testing ground because Estonia has long been viewed as a benchmark for digital government. The country is known for its national digital identity infrastructure, online public services, and a policy culture that treats digital access as part of the state’s operating model rather than an add-on.
StepHow’s PoC will sit inside that environment, connecting Wissly to three Tallinn information systems. The first is Teele, the city’s legal acts information system, used for drafting and processing legal acts, resolutions, and directives from the Tallinn City Council, City Government, and related authorities.
The second is AKIS, the public gatherings information system, which manages public events and gathering permit processes in Tallinn. The third is Postipoiss, the city’s document management system, which stores and processes major administrative documents and correspondence, including the 2026 contracts register.
StepHow delivered a working demo of Wissly to the Tallinn project team on June 23 and launched the testing environment. The assistant works on both desktop and mobile devices. Internal users can ask questions in Estonian or English, after which the system searches connected administrative data in real time and generates answers using what the company describes as a “citation-first” approach.
In practice, that means every claim in an answer is tied to a source link from the original document. This is StepHow’s attempt to address one of the biggest weaknesses of large language models: their tendency to produce plausible-sounding but unsupported answers.
Why citations are the core feature
For consumer chatbots, a vague answer can be irritating. In public administration, it can create legal and operational problems.
Officials handling permits, contracts, legal acts, or public correspondence need to know not just what a system says, but why it says it. If an AI assistant summarises the wrong version of a document, misses a later amendment, or cites an irrelevant source, the cost may be borne by citizens, businesses, or other departments.
Also Read: Can your AI actually read your data?
That is why the Tallinn PoC is limited to designated internal staff and uses only publicly available public data and PDF documents. The restricted scope reduces risk while allowing the city and StepHow to test whether the system can cope with real administrative workflows.
The collaboration is scheduled to run until November 30, 2026. Depending on the results, the two sides may discuss a citizen-facing service expansion, further pilots, or a commercial deployment.
“We are thrilled to undertake this innovative validation agreement with the City of Tallinn, a global benchmark for digital governance,” said Hwang. “Through Wissly, which enables question-answering and rigorous evidence verification for public records, we will maximise administrative efficiency and ultimately demonstrate the welfare benefits of innovating how citizens access public information.”
What Southeast Asia can take from the experiment
Although the project is in Europe, its relevance extends to Southeast Asia, where governments are also pushing digital public services but often face fragmented back-end systems.
Singapore has been early in adopting digital identity, data exchange frameworks, and government technology platforms. Indonesia, the Philippines, Vietnam, Malaysia, and Thailand are also investing in digital government services, though implementation varies widely between national agencies, local governments, and legacy departments.
The problem StepHow is addressing is familiar across the region: information exists, but it is scattered across portals, PDFs, old databases, and internal records. A citizen may want to understand a permit requirement. A civil servant may need to confirm whether a policy is still valid. A business may need clarity on licensing or procurement rules. In many cases, the bottleneck is not the absence of information, but the difficulty of retrieving and verifying it quickly.
That is where AI search could be useful, especially in multilingual societies. Southeast Asia’s public sector operates across languages, dialects, and varying levels of digital literacy. But the region also presents a harder challenge than a controlled internal PoC. Citizen-facing AI in markets such as Indonesia or the Philippines would need to handle local languages, uneven document quality, inconsistent data structures, and politically sensitive public information.
Tallinn’s test therefore offers a useful but limited reference point. It shows how a city with mature digital infrastructure might trial AI safely before public rollout. For Southeast Asian governments, the larger lesson may be procedural: start with bounded use cases, require citations, restrict sensitive data, and test systems with officials before exposing them to the public.
A crowded field for enterprise AI search
StepHow is entering a competitive space. Enterprise AI search and workplace assistants have become a major battleground since the rise of generative AI. Globally, players such as Glean, Microsoft Copilot, Google Vertex AI Search, OpenAI’s enterprise products, and AWS-backed AI services are all trying to help organisations search internal knowledge and automate routine information work.
In the public sector, StepHow may also run into competition from large systems integrators and govtech vendors that already manage government document platforms, identity systems, and cloud infrastructure. Its differentiation, based on the Tallinn project, appears to be a focus on citation-heavy administrative search rather than a general-purpose productivity assistant.
For a smaller AI company, that specialisation can be an advantage if it proves reliable. Public agencies are unlikely to adopt tools that cannot show their sources. But selling into government also means long procurement cycles, security reviews, localisation demands, and accountability requirements that are more demanding than typical enterprise software sales.
The Tallinn PoC will not, by itself, prove that Wissly can scale across governments or markets. It will, however, test the product in a demanding environment where document accuracy and institutional trust are central.
Also Read: Singapore’s data analysts trust AI to work, not to think
If StepHow can show that its assistant reduces the time officials spend searching records without sacrificing reliability, the company may gain a useful reference case for Europe and, potentially, for digital government projects in Asia.
For now, the real story is not that Tallinn is testing another chatbot. It is that the next phase of AI in government may be less about flashy automation and more about something less glamorous but far more important: helping public servants find the right answer, and prove it.
The post Tallinn taps StepHow to test AI assistant for public-sector document search appeared first on e27.