diff --git a/milli/src/update/index_documents/extract/extract_geo_points.rs b/milli/src/update/index_documents/extract/extract_geo_points.rs
index 0a73e5ed4..1849d5f5d 100644
--- a/milli/src/update/index_documents/extract/extract_geo_points.rs
+++ b/milli/src/update/index_documents/extract/extract_geo_points.rs
@@ -21,19 +21,22 @@ pub fn extract_geo_points<R: io::Read>(
     })?;
 
     // we never encountered any documents with a `_geo` field. We can skip entirely this step
-    if geo_field_id.is_none() {
-        return Ok(writer_into_reader(writer)?);
-    }
-    let geo_field_id = geo_field_id.unwrap();
+    let geo_field_id = match geo_field_id {
+        Some(geo) => geo,
+        None => return Ok(writer_into_reader(writer)?),
+    };
 
     while let Some((docid_bytes, value)) = obkv_documents.next()? {
         let obkv = obkv::KvReader::new(value);
-        let point = obkv.get(geo_field_id).unwrap(); // TODO: TAMO where should we handle this error?
+        let point = match obkv.get(geo_field_id) {
+            Some(point) => point,
+            None => continue,
+        };
         let point: Value = serde_json::from_slice(point).map_err(InternalError::SerdeJson)?;
 
         if let Some((lat, lng)) = point["lat"].as_f64().zip(point["lng"].as_f64()) {
             // this will create an array of 16 bytes (two 8 bytes floats)
-            let bytes: [u8; 16] = concat_arrays![lat.to_le_bytes(), lng.to_le_bytes()];
+            let bytes: [u8; 16] = concat_arrays![lat.to_ne_bytes(), lng.to_ne_bytes()];
             writer.insert(docid_bytes, bytes)?;
         } else {
             // TAMO: improve the warn
diff --git a/milli/src/update/index_documents/extract/mod.rs b/milli/src/update/index_documents/extract/mod.rs
index 90a279815..736060b15 100644
--- a/milli/src/update/index_documents/extract/mod.rs
+++ b/milli/src/update/index_documents/extract/mod.rs
@@ -51,13 +51,14 @@ pub(crate) fn data_from_obkv_documents(
                 lmdb_writer_sx.clone(),
                 &searchable_fields,
                 &faceted_fields,
+                geo_field_id,
                 &stop_words,
             )
         })
         .collect();
 
     let (
-        (docid_word_positions_chunks),
+        docid_word_positions_chunks,
         (docid_fid_facet_numbers_chunks, docid_fid_facet_strings_chunks),
     ) = result?;
 
@@ -121,16 +122,6 @@ pub(crate) fn data_from_obkv_documents(
         "field-id-facet-number-docids",
     );
 
-    spawn_extraction_task(
-        documents_chunk,
-        indexer.clone(),
-        lmdb_writer_sx.clone(),
-        move |documents, indexer| extract_geo_points(documents, indexer, geo_field_id),
-        merge_cbo_roaring_bitmaps,
-        TypedChunk::GeoPoints,
-        "geo-points",
-    );
-
     Ok(())
 }
 
@@ -181,6 +172,7 @@ fn extract_documents_data(
     lmdb_writer_sx: Sender<Result<TypedChunk>>,
     searchable_fields: &Option<HashSet<FieldId>>,
     faceted_fields: &HashSet<FieldId>,
+    geo_field_id: Option<FieldId>,
     stop_words: &Option<fst::Set<&[u8]>>,
 ) -> Result<(
     grenad::Reader<CursorClonableMmap>,
@@ -190,6 +182,12 @@ fn extract_documents_data(
 
     let _ = lmdb_writer_sx.send(Ok(TypedChunk::Documents(documents_chunk.clone())));
 
+    let (documents_chunk_cloned, lmdb_writer_sx_cloned) = (documents_chunk.clone(), lmdb_writer_sx.clone());
+    rayon::spawn(move || {
+        let geo_points = extract_geo_points(documents_chunk_cloned, indexer, geo_field_id).unwrap();
+        lmdb_writer_sx_cloned.send(Ok(TypedChunk::GeoPoints(geo_points))).unwrap();
+    });
+
     let (docid_word_positions_chunk, docid_fid_facet_values_chunks): (Result<_>, Result<_>) =
         rayon::join(
             || {
diff --git a/milli/src/update/index_documents/typed_chunk.rs b/milli/src/update/index_documents/typed_chunk.rs
index 0dfeabece..9605fea7d 100644
--- a/milli/src/update/index_documents/typed_chunk.rs
+++ b/milli/src/update/index_documents/typed_chunk.rs
@@ -181,16 +181,16 @@ pub(crate) fn write_typed_chunk_into_index(
         }
         TypedChunk::GeoPoints(mut geo_points) => {
             // TODO: TAMO: we should create the rtree with the `RTree::bulk_load` function
-            let mut rtree = index.geo_rtree(&index.read_txn()?)?.unwrap_or_default();
+            let mut rtree = index.geo_rtree(wtxn)?.unwrap_or_default();
             while let Some((key, value)) = geo_points.next()? {
                 // convert the key back to a u32 (4 bytes)
                 let (key, _) = helpers::try_split_array_at::<u8, 4>(key).unwrap();
-                let key = u32::from_le_bytes(key);
+                let key = u32::from_be_bytes(key);
 
                 // convert the latitude and longitude back to a f64 (8 bytes)
                 let (lat, tail) = helpers::try_split_array_at::<u8, 8>(value).unwrap();
                 let (lng, _) = helpers::try_split_array_at::<u8, 8>(tail).unwrap();
-                let point = [f64::from_le_bytes(lat), f64::from_le_bytes(lng)];
+                let point = [f64::from_ne_bytes(lat), f64::from_ne_bytes(lng)];
                 rtree.insert(GeoPoint::new(point, key));
             }
             index.put_geo_rtree(wtxn, &rtree)?;