Context. We explore the stellar content of the Javalambre Photometric Local Universe Survey (J-PLUS) Data Release 2 and show its potential for identifying low-metallicity stars using the Stellar Parameters Estimation based on Ensemble Methods (SPEEM) pipeline. Aims. SPEEM is a tool used to provide determinations of atmospheric parameters for stars and separate stellar sources from quasars based on the unique J-PLUS photometric system. The adoption of adequate selection criteria allows for the identification of metal-poor star candidates that are suitable for spectroscopic follow-up investigations. Methods. SPEEM consists of a series of machine-learning models that use a training sample observed by both J-PLUS and the SEGUE spectroscopic survey. The training sample has temperatures, T eff , between 4800 K and 9000 K, values of log g between 1.0 and 4.5, as well as −3.1 < Fe/H < +0.5. The performance of the pipeline was tested with a sample of stars observed by the LAMOST survey within the same parameter range. Results. The average differences between the parameters of a sample of stars observed with SEGUE and J-PLUS, obtained with the SEGUE Stellar Parameter Pipeline and SPEEM, respectively, are Δ T eff ~ 41 K, Δlog g ~ 0.11 dex, and ΔFe/H ~ 0.09 dex. We define a sample of 177 stars that have been identified as new candidates with Fe/H < −2.5, with 11 of them having been observed with the ISIS spectrograph at the William Herschel Telescope. The spectroscopic analysis confirms that 64% of stars have Fe/H < −2.5, including one new star with Fe/H < −3.0. Conclusions. Using SPEEM in combination with the J-PLUS filter system has demonstrated their potential in estimating the stellar atmospheric parameters ( T eff , log g , and Fe/H). The spectroscopic validation of the candidates shows that SPEEM yields a success rate of 64% on the identification of very metal-poor star candidates with Fe/H < −2.5.