Projects / AI for 30 years of research
Unlocking 30 years of research with AI
Helping Svenskt Vatten instantly unlock 50 000 pages of research.
“Lens revolutionizes how I find answers. Our 800+ reports are now instantly searchable, even when I'm on the run with my phone.”
Ulf Thysell
Strategy, Svenskt Vatten
Keywords
AI
Knowledge bases
Document understanding
LLMs
Retrieval-Augmented Generation (RAG)
UX / UI
Introduction
Svenskt Vatten is Sweden's industry organization for water and wastewater utilities, representing over 300 member organizations that collectively serve the entire Swedish population.
Through its research arm, Svenskt Vatten Utveckling (SVU), the organization has accumulated over 800 lenghty research reports spanning three decades. A comprehensive knowledge base covering everything from water tariff models to PFAS remediation costs.
Challenge
With 800+ reports and counting, the SVU library had become a victim of its own success. With individual reports ranging from 50 to 300 pages, the archive had grown to over 50 000 pages.
Researchers, leaders, and officials struggled to effectively navigate the archive using traditional search tools. The most telling symptom: new research applications were repeatedly being submitted for topics that had already been studied. Existing search tools lacked the precision to surface relevant prior work, and no one had the bandwidth to manually cross-reference the full library.
General-purpose AI tools weren't the answer either, they couldn't handle the sheer size of the archive, and also lacked the domain-specific knowledge and accuracy that water sector professionals required.
Goal
Make 30+ years of SVU research instantly accessible, searchable, and useful.
Internally, the tool would reduce duplicate research applications by surfacing existing work earlier in the process. Externally, it would empower members to self-serve answers directly from the report library ultimately reducing the volume of inbound questions to Svenskt Vatten on topics already well-documented in SVU's archive.
Solution
Vattensnack, a specialized RAG-based (Retrieval-Augmented Generation) AI platform trained on the complete SVU report library. The tool allows users to have a natural conversation with the entire archive by asking questions in plain language and receiving fact-based answers with source references directly to the underlying reports.
The system was fine-tuned with sector-specific terminology, including the full dictionary of Swedish water and wastewater industry abbreviations. Unlike generic AI, Vattensnack is scoped strictly to the SVU corpus, which keeps answers grounded and accurate.
The real challenge
Early in the process, we identified that the real challenge was user trust and adoption. Personal AI tools that people relied on in their private lives consistently underperformed in professional settings, and when users couldn't verify an answer, they stopped using the tool.
Vattensnack was built around this insight: every response includes exact references to the specific report sections the answer was drawn from. Users stay in control, answers are verifiable, and adoption follows naturally from transparency.
Launch
Following successful internal testing and a review period with the SVU committee, Vattensnack was opened to a broader pilot in January 2026 with over 150 users.
Results
Early feedback from the pilot confirmed the core value proposition. Users could find and filter information quickly, with source references that made it easy to validate answers and locate the original reports.
Users could now surface relevant research in seconds rather than hours, saving significant time on literature reviews and reducing the risk of overlooking prior work. Members who previously had to contact Svenskt Vatten directly with questions, could use the tool as an immediate, reliable alternative for many questions.
As Ulf Thysell, Climate Adaptation and Civil Preparedness Strategist at Svenskt Vatten, put it: "It's easy to find and sort information, you eliminate the risk of duplicate work, and you get fact-based results with genuinely high accuracy."