LAaDT-DSS pairs rooftop solar decisions with a language model that only calls the tools
The decision-support system answers investment, incentive and grid-limit questions for non-experts under Turkish tariff rules, using one simulation core and a language model restricted to tool calls.
In plain words
Researchers built software that helps people without modelling skills decide whether rooftop solar panels make sense for a home. It answers three linked questions at once: whether to invest, which government incentive costs the state least, and whether the local grid stays within its limits. A language model acts only as an interface that triggers the simulations, and every input value is labelled as sourced or estimated.
What was announced
The team behind LAaDT-DSS, described in an arXiv abstract, says users reach the system by clicking a house on a 3D map or by asking a question in plain words. The language model is asked to call the simulation tools rather than compute results itself, and the system's 125-entry parameter registry marks each entry as sourced or estimated. In case studies, the reported photovoltaic yield fell within ±3.6% of two independent tools at equal losses.
Robustness testing drew on 2,048 random draws of the estimated inputs. The answer for photovoltaics alone held in all but one draw, while adding a battery changed the answer in 31–41% of draws. Four scripted chat scenarios ran end to end, but the authors report that no user study was conducted, so real-world usability remains untested.
What it does not show
- The evidence is the paper's abstract only; the full article was not reviewed, and no user study was run, so usability with real non-experts is untested.
- Reported accuracy and robustness figures come from the authors' own case studies and sampling; no independent confirmation is available.
For security and IT teams
- Scope
- The system is built around Turkish tariff rules, so its answers and parameter registry reflect that regulatory setting rather than other markets.
Sources
Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 20 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.