MetInArt
Asistent pentru crearea de prognoze meteorologice cu Large Language Models (LLM)
We at SINAPSA Infocomplex®™
have used our knowledge of meteorology, software development and Artificial Intelligence
and have created this program, with the aim to provide the weather forecasters
with a tool that would streamline the production of various forecasts as complex text,
and to gain time for their other tasks.
The first such program ever made in Europe,
with us at SINAPSA Infocomplex®™ being the first private-sector enterprise
who have attempted and succeeded in such a feat.
The forecast for a specific place (city, or a very small region)
can be aided by the relevant information for a larger area where that place exists –
that is, for example, information relevant for the entirety of a country
can be stored ONCE and reused for creating the prompts for the forecasts
for cities or small regions in that country.
The capabilities of this program are dependent on what the requirements are exactly,
from the LLM that has to process the final prompt and provide forecasts.
The program can ask the LLM to perform these procedures:
- build the prompt aimed at creating a forecast...
... by using the human operator-provided information at the time of operation
... by using the information that the LLM has been trained on
- use tools
- use RAG
What can MetInArt do:
- build a prompt in an organized way, that will be sent to an LLM which in turn will create various forecasts
- adjust the level of detail in the forecast: very concise, concise, detailed, very detailed;
- adjust the style of the forecast: continuous text, or sketchy/tabelar format;
- adjust the prompt as to demand the forecast of – or, the risk/probability for: fog, frost, ice, heavy rain, flash floods, hail, wind gusts, thunderstorms, snowdrifts, high waves, landslides, traffic conditions;
- build the prompt as to demand the estimation of favorable conditions for astronomical observations from the ground, and for open-air events;
- build the prompt as to ask for the "personal" opinion of the LLM regarding a situation;
- build the prompt as for the LLM to use precise information about:
- precise location and general conditions in the place of interest: altitude, surrounding landscape, bodies of water, distance to mountain peaks, etc.
- isobars, and air circulation at ground level
- isotherms at various standard levels in the upper atmosphere (850 hPa, 700 hPa, etc.)
- isohypses at various standard levels in the upper atmosphere (850 hPa, 700 hPa, etc.)
- air temperature at ground level: maximum and minimum values, their location
- atmospheric fronts
- distances to major atmospheric features
- inception of changes in major atmospheric features
- duration of changes in major atmospheric features
- estimated time-of-arrival of fronts and isotherms
- the initial, overall situation (air masses, pressure systems, current and past weather, etc.)
- allow for any piece of information to be changed before sending the prompt to the LLM
- allow for changing the prompt before sending it to the LLM
- store detailed information about particular places in a country or more (regions, cities, other settlements);
- save the information about a place, provided by the human operator, for repeated/later use in creating the forecast for that placs; for example, daily forecasts for a county or a mountain resort;
- change that information, in order to increase its level of detail, and save/use it again;
- establish the language of the forecast built by the LLM – that is, write the forecast directly in that language, without the need of using a translator after the original composition; in the October 2025 version of the program, the user can select from these languages: Romanian, Deutsch, English, Français, Italiano, Russian
SINAPSA Infocomplex®™