EUFUNDS

LLM-Forecaster - From Data Streams to reliable forecasts

Total project value:

PLN 9,688,091.11

Grant Value:

PLN 4,352,916.87

CONTEXT AND CHALLENGE

Forecasting is no longer limited to static statistical models.

Organisations operating on large, continuously changing datasets need predictions that keep pace with real-time signals, adapt to new information, and support confident decision-making across multiple domains. Financial markets, logistics networks, sales pipelines, and transport systems all rely on accurate forecasts, yet traditional approaches struggle with scale, complexity, and anomaly detection.

LLM-Forecaster was initiated to address these limitations through an advanced, automated forecasting system built on Large Language Models and Retrieval-Augmented Generation.

See results and impact

Project objective

LLM-Forecaster is structured as an R&D project consisting of three complementary stages, covering the full path from research to deployment.

Scope and approach

The goal of the project is to develop a unique technical solution that enables fast, precise, and automated prediction of events based on large time-series datasets. By combining LLM-based reasoning with real-time data processing, the project aims to move forecasting beyond reactive analysis and towards adaptive, intelligence-driven prediction.

Industrial Research

Data processing and preparation

The first stage focuses on building a module for automated collection, processing, and selection of data from multiple real-time sources. This work resulted in optimised datasets tailored specifically for time-series forecasting, improving both data quality and model performance. A key outcome of this phase is higher prediction accuracy of LLM-based models and the development of effective methods for anomaly detection.

Forecasting methods and algorithms

The second stage concentrates on the design and evaluation of forecasting methods based on LLM architectures. This research leads to the creation of a dedicated LLM-Forecaster algorithm, increasing predictive accuracy across key business sectors, including finance, transport, sales, and logistics.

Experimental Development

System integration and validation

The final stage integrates the results of previous phases into a complete forecasting module. This includes user interfaces and API access, enabling practical use at scale. The outcome is a fully functional and validated LLM-Forecaster system, tested with an external client and ready for large-scale data processing and clear presentation of forecasting results.

Target users

The solution is designed for:

Companies operating in finance, trade, logistics, and transport,
Enterprises using data analysis and AI solutions,
B2B organisations seeking tools for prediction and process optimisation.

Results and impact

The project delivers:

New forecasting methods and algorithms based on LLM and RAG

Deployment-ready LLM-Forecaster system

Implementation of the solution into the applicant’s business operations within six months of project completion

LLM-Forecaster transforms time-series forecasting into a practical, automated capability that supports faster decisions and higher confidence in environments driven by large and complex datasets.

Curious about how we can DELIVER in your project?