diff --git a/talks/manu_parra.md b/talks/manu_parra.md
new file mode 100644
index 0000000..df332d7
--- /dev/null
+++ b/talks/manu_parra.md
@@ -0,0 +1,30 @@
+# Neutral hydrogen profile classification using Convolutional Neural Networks
+
+## Authors
+
+Manuel Parra-Royón (IAA-CSIC), Laura Darriba-Pol (IAA-CSIC) and Lourdes Verdes-Montenegro (IAA-CSIC)
+
+## Presenter
+
+Manuel Parra-Royón
+
+## Abstract (200 words max.) 
+
+HI (not Hello, not Hi, just Neutral Hydrogen), also known as the Cosmic Hydrogen Cool Kids Club 😀, is basically the 'cool' version of hydrogen floating in space. Picture it: it's just like normal hydrogen, but it's not in the mood for wild parties, so it just sits there, emitting this distinctive 21-centimetre wavelength signal as if it were the universal radio streaming station. Radio astronomers love this "HI radio" because it tells them how hydrogen lives in galaxies ;). It's the ultimate party pooper, giving us the scoop on how galaxies mingle, merge and sometimes even throw epic cosmic parties. Not to mention that it's the VIP pass to the dark matter disco, allowing us to dance to the gravitational beat of the universe. In a few words, HI is the celestial DJ that rocks the galactic dance floor and informs us on everything from cosmic makeovers to gravitational dance moves, rotation curves and/or/to galactic structures and evolution.
+
+In this context, HI classification is important as it enables astronomers to discern and categorize the diverse properties and behaviors of neutral hydrogen gas within galaxies, providing crucial insights into galactic structures, interactions, and evolution. We want to present our work in progress on how to refine this type of profiling for classification using different approaches with Deep Neural Networks, in particular exploring how Convolutional Neural Networks (CNN) can help this. The complexity in applying Convolutional Neural Networks (CNNs) to HI profiling lies in the intricate nature of the HI data and the nuanced spatial dependencies within it. HI profiles often exhibit complex patterns and variations. Furthermore, the structural diversity and subtle features within the HI data demand a CNN architecture that can efficiently extract relevant information while minimizing overfitting. Balancing the complexity of the model with the need for generalisation is key to ensuring the effectiveness of the CNN in accurately classifying the various HI profiles.
+
+In terms of data (Big Big Data), we will use available data from Arecibo and Very Large Array (VLA) catalogues principally, aiming to construct and verify a minimum methodology that could potentially be applied to ongoing surveys of Square Kilometer Array (SKA) precursors such as MeerKAT (MIGHTEE HI) or Apertif.
+
+
+## References
+
+- https://www.sciencedirect.com/science/article/pii/S1387647306002892
+- https://iopscience.iop.org/article/10.3847/1538-4357/ac5ea0/pdf
+- VLA: https://en.wikipedia.org/wiki/Very_Large_Array
+- SKA and SKAO: https://www.skao.int/
+- MeerKAT: http://mgcls.sarao.ac.za/
+- Apertif: https://www.astron.nl/telescopes/wsrt-apertif/
+
+
+