diff --git a/crates/lib/mysql_queries/src/queries/database_time/get_database_time.rs b/crates/lib/mysql_queries/src/queries/database_time/get_database_time.rs
new file mode 100644
index 0000000000..57a2817bbe
--- /dev/null
+++ b/crates/lib/mysql_queries/src/queries/database_time/get_database_time.rs
@@ -0,0 +1,30 @@
+use anyhow::anyhow;
+use chrono::{DateTime, Utc};
+use sqlx::MySqlPool;
+
+use errors::AnyhowResult;
+
+pub struct GetDatabaseTimeResult {
+  pub database_time: DateTime<Utc>,
+}
+
+/// Query the DB for its current time
+pub async fn get_database_time(
+  pool: &MySqlPool,
+) -> AnyhowResult<GetDatabaseTimeResult> {
+
+  let maybe_record = sqlx::query_as!(
+      GetDatabaseTimeResult,
+        r#"
+SELECT NOW() as `database_time: DateTime<Utc>`
+        "#,
+    )
+      .fetch_one(pool)
+      .await;
+
+  match maybe_record {
+    Ok(record) => Ok(record),
+    Err(sqlx::Error::RowNotFound) => Err(anyhow!("query didn't return anything")),
+    Err(ref err) => Err(anyhow!("database query error: {:?}", err)),
+  }
+}
diff --git a/crates/lib/mysql_queries/src/queries/database_time/mod.rs b/crates/lib/mysql_queries/src/queries/database_time/mod.rs
new file mode 100644
index 0000000000..43a9db71b5
--- /dev/null
+++ b/crates/lib/mysql_queries/src/queries/database_time/mod.rs
@@ -0,0 +1 @@
+pub mod get_database_time;
\ No newline at end of file
diff --git a/crates/lib/mysql_queries/src/queries/mod.rs b/crates/lib/mysql_queries/src/queries/mod.rs
index 46e996096c..cfc3208f40 100644
--- a/crates/lib/mysql_queries/src/queries/mod.rs
+++ b/crates/lib/mysql_queries/src/queries/mod.rs
@@ -5,6 +5,7 @@ pub mod beta_keys;
 pub mod billing;
 pub mod browser_session_logs;
 pub mod comments;
+pub mod database_time;
 pub mod email_sender_jobs;
 pub mod entity_stats;
 pub mod featured_items;
diff --git a/crates/service/job/es-update-job/src/main.rs b/crates/service/job/es-update-job/src/main.rs
index de6c0dd682..95a2c71993 100644
--- a/crates/service/job/es-update-job/src/main.rs
+++ b/crates/service/job/es-update-job/src/main.rs
@@ -17,6 +17,7 @@
 //#![forbid(warnings)]
 
 use std::sync::Arc;
+
 use elasticsearch::Elasticsearch;
 use elasticsearch::http::transport::Transport;
 use log::info;
@@ -29,8 +30,10 @@ use config::shared_constants::DEFAULT_RUST_LOG;
 use errors::AnyhowResult;
 
 use crate::job_state::{JobState, SleepConfigs};
-use crate::tasks::update_all_model_weights::update_all_model_weights::update_all_model_weights;
+use crate::tasks::model_weights::update_all_model_weights::update_all_model_weights;
+use crate::tasks::model_weights::update_recently_written_model_weights::update_recently_written_model_weights;
 use crate::tasks::update_engine_media_files::update_engine_media_files::update_engine_media_files;
+
 pub mod job_state;
 pub mod tasks;
 
@@ -60,23 +63,35 @@ async fn main() -> AnyhowResult<()> {
     },
   });
 
-  info!("Starting thread to update all model weights...");
+  info!("Starting thread to update recently updated model weights...");
 
   let job_state_1 = job_state.clone();
 
   let handle_1 = tokio::task::spawn(async move {
-    let _r = update_all_model_weights(job_state_1).await;
+    let _r = update_recently_written_model_weights(job_state_1).await;
   });
 
-  info!("Starting thread to update engine media files...");
+  info!("Starting thread to update all model weights...");
 
   let job_state_2 = job_state.clone();
 
   let handle_2 = tokio::task::spawn(async move {
-    let _r = update_engine_media_files(job_state_2).await;
+    let _r = update_all_model_weights(job_state_2).await;
   });
 
-  futures::future::join_all([handle_1, handle_2]).await;
+  info!("Starting thread to update engine media files...");
+
+  let job_state_3 = job_state.clone();
+
+  let handle_3 = tokio::task::spawn(async move {
+    let _r = update_engine_media_files(job_state_3).await;
+  });
+
+  futures::future::join_all([
+    handle_1,
+    handle_2,
+    handle_3,
+  ]).await;
 
   Ok(())
 }
diff --git a/crates/service/job/es-update-job/src/tasks/mod.rs b/crates/service/job/es-update-job/src/tasks/mod.rs
index 9e5e07fa87..098e60a426 100644
--- a/crates/service/job/es-update-job/src/tasks/mod.rs
+++ b/crates/service/job/es-update-job/src/tasks/mod.rs
@@ -1,2 +1,2 @@
-pub mod update_all_model_weights;
+pub mod model_weights;
 pub mod update_engine_media_files;
\ No newline at end of file
diff --git a/crates/service/job/es-update-job/src/tasks/model_weights/mod.rs b/crates/service/job/es-update-job/src/tasks/model_weights/mod.rs
new file mode 100644
index 0000000000..7a4511dcae
--- /dev/null
+++ b/crates/service/job/es-update-job/src/tasks/model_weights/mod.rs
@@ -0,0 +1,3 @@
+pub mod update_all_model_weights;
+pub mod update_recently_written_model_weights;
+pub mod util;
\ No newline at end of file
diff --git a/crates/service/job/es-update-job/src/tasks/model_weights/update_all_model_weights.rs b/crates/service/job/es-update-job/src/tasks/model_weights/update_all_model_weights.rs
new file mode 100644
index 0000000000..f1a508cca5
--- /dev/null
+++ b/crates/service/job/es-update-job/src/tasks/model_weights/update_all_model_weights.rs
@@ -0,0 +1,100 @@
+use std::sync::Arc;
+use std::time::Duration;
+
+use chrono::{DateTime, Utc};
+use log::{error, info};
+
+use errors::AnyhowResult;
+use mysql_queries::queries::model_weights::batch_get::batch_get_model_weights_for_elastic_search_backfill::batch_get_model_weights_for_elastic_search_backfill;
+use mysql_queries::queries::model_weights::list::list_model_weight_tokens_updated_since::list_model_weight_tokens_updated_since;
+
+use crate::job_state::JobState;
+use crate::tasks::model_weights::util::create_model_weight_document_from_record::create_model_weight_document_from_record;
+
+pub async fn update_all_model_weights(job_state: Arc<JobState>) {
+  // TODO(bt,2024-02-05): Write this cursor to Redis so job can resume without reindexing everything.
+  let mut cursor = DateTime::UNIX_EPOCH;
+
+  loop {
+    info!("Main loop; cursor @ {:?}", &cursor);
+
+    let result = with_database_main_loop(&mut cursor, &job_state).await;
+
+    if let Err(err) = result {
+      error!("Error in main loop: {:?}", err);
+      tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_error_wait_millis)).await;
+    }
+
+    // NB: Function should never return in success case, but we'll sleep just in case.
+  }
+}
+
+pub async fn with_database_main_loop(updated_at_cursor: &mut DateTime<Utc>, job_state: &JobState) -> AnyhowResult<()> {
+  info!("Acquiring MySQL connection...");
+
+  let mut mysql_connection = job_state.mysql_pool.acquire().await?;
+
+  let mut last_successful_update_at = *updated_at_cursor;
+
+  loop {
+    info!("Querying tokens updated since: {:?}", &updated_at_cursor);
+
+    let mut maybe_tokens =
+        list_model_weight_tokens_updated_since(&mut *mysql_connection, &updated_at_cursor).await?;
+
+    if maybe_tokens.is_empty() {
+      info!("No records updated since {:?}", &updated_at_cursor);
+      tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_no_updates_wait_millis)).await;
+      continue;
+    }
+
+    info!("Found {} updated records", maybe_tokens.len());
+
+    let mut last_observed_updated_at = *updated_at_cursor;
+
+    while !maybe_tokens.is_empty() {
+      // NB: This list might be very large if we query from (1) the epoch, or (2) there was a large series of updates
+      let last = 50.min(maybe_tokens.len());
+      let drained_tokens = maybe_tokens.drain(0..last)
+          .into_iter()
+          .map(|record| record.token)
+          .collect::<Vec<_>>();
+
+      let records
+          = batch_get_model_weights_for_elastic_search_backfill(&mut *mysql_connection, &drained_tokens).await?;
+
+      for record in records {
+        let updated_at = record.updated_at.clone();
+
+        create_model_weight_document_from_record(&job_state.elasticsearch, record).await?;
+
+        // NB: We don't want to advance the cursor to the current second just yet, because we might be processing
+        // several records with the exact same "updated_at" timestamp. We do know that due to ordering, we can update
+        // the timestamp cursor to the second before this timestamp, however.
+        //
+        // If we change the cursoring or batch sizes, we'll need to reconsider all of this logic.
+        if let Some(cursor_at_least) = updated_at.checked_sub_signed(chrono::Duration::seconds(1)) {
+          // NB: We don't want the cursor to slide backwards (it shouldn't, but the max will save us).
+          last_successful_update_at = cursor_at_least.max(last_successful_update_at);
+        }
+
+        *updated_at_cursor = last_successful_update_at.max(*updated_at_cursor);
+
+        last_observed_updated_at = updated_at;
+
+        tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_es_writes_wait_millis)).await;
+      }
+
+      tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_job_batch_wait_millis)).await;
+    }
+
+    // NB: The last cursor math put us a second behind the current clock. Here we'll advance to the last record.
+    // NB: This technically could miss records if we're updating within the same second batch we read from, but that
+    // seems both unlikely and not worth solving at our present scale.
+    *updated_at_cursor = last_observed_updated_at.max(*updated_at_cursor);
+
+    info!("Up to date at cursor = {:?}", &updated_at_cursor);
+
+    tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_query_wait_millis)).await;
+  }
+}
diff --git a/crates/service/job/es-update-job/src/tasks/model_weights/update_recently_written_model_weights.rs b/crates/service/job/es-update-job/src/tasks/model_weights/update_recently_written_model_weights.rs
new file mode 100644
index 0000000000..c5e6ee09d4
--- /dev/null
+++ b/crates/service/job/es-update-job/src/tasks/model_weights/update_recently_written_model_weights.rs
@@ -0,0 +1,91 @@
+use std::ops::Add;
+use std::sync::Arc;
+use std::time::Duration;
+
+use chrono::{DateTime, Utc};
+use log::{error, info};
+
+use errors::AnyhowResult;
+use mysql_queries::queries::database_time::get_database_time::get_database_time;
+use mysql_queries::queries::model_weights::list::list_model_weight_tokens_updated_since::list_model_weight_tokens_updated_since;
+
+use crate::job_state::JobState;
+use crate::tasks::model_weights::util::copy_model_weight_records_to_documents::copy_model_weight_records_to_documents;
+
+pub async fn update_recently_written_model_weights(job_state: Arc<JobState>) {
+  let mut last_updated_time= DateTime::UNIX_EPOCH;
+
+  // TODO(bt,2024-02-05): Write this cursor to Redis so job can resume without reindexing everything.
+  loop {
+    match get_database_time(&job_state.mysql_pool).await {
+      Ok(db_time) => {
+        last_updated_time = db_time.database_time;
+        break;
+      },
+      Err(err) => {
+        error!("Error getting database time: {:?}", err);
+        tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_error_wait_millis)).await;
+        continue;
+      }
+    }
+  }
+
+  loop {
+    info!("Main loop; cursor @ {:?}", &last_updated_time);
+    let result = with_database_main_loop(&mut last_updated_time, &job_state).await;
+
+    if let Err(err) = result {
+      error!("Error in main loop: {:?}", err);
+      tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_error_wait_millis)).await;
+    }
+
+    // NB: Function should never return in success case, but we'll sleep just in case.
+  }
+}
+
+pub async fn with_database_main_loop(updated_at_cursor: &mut DateTime<Utc>, job_state: &JobState) -> AnyhowResult<()> {
+  info!("Acquiring MySQL connection...");
+
+  let mut mysql_connection = job_state.mysql_pool.acquire().await?;
+
+  loop {
+    info!("Querying tokens updated since: {:?}", &updated_at_cursor);
+
+    let maybe_records =
+        list_model_weight_tokens_updated_since(&mut *mysql_connection, &updated_at_cursor).await?;
+
+    if maybe_records.is_empty() {
+      info!("No records updated since {:?}", &updated_at_cursor);
+      tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_no_updates_wait_millis)).await;
+      continue;
+    }
+
+    let maybe_tokens = maybe_records.into_iter()
+        .map(|record| record.token)
+        .collect::<Vec<_>>();
+
+    info!("Found {} updated records", maybe_tokens.len());
+
+    let maybe_last_cursor = copy_model_weight_records_to_documents(
+      maybe_tokens,
+      &mut *mysql_connection,
+      &job_state.elasticsearch,
+      &job_state.sleep_config,
+    ).await?;
+
+    if let Some(last_cursor) = maybe_last_cursor {
+      // NB: If we trigger thousands of records to get updated at the same time, the cursor could get stuck.
+      // This is a simple way to ensure we don't get stuck, but we could miss out on some updates.
+      let delta = last_cursor.updated_at.signed_duration_since(*updated_at_cursor);
+
+      if delta.num_seconds() == 0 {
+        // Cursor didn't move, so we need to advance it.
+        *updated_at_cursor = last_cursor.updated_at.add(Duration::from_secs(1));
+      } else {
+        *updated_at_cursor = last_cursor.updated_at;
+      }
+    }
+
+    tokio::time::sleep(Duration::from_millis(job_state.sleep_config.between_query_wait_millis)).await;
+  }
+}
diff --git a/crates/service/job/es-update-job/src/tasks/model_weights/util/copy_model_weight_records_to_documents.rs b/crates/service/job/es-update-job/src/tasks/model_weights/util/copy_model_weight_records_to_documents.rs
new file mode 100644
index 0000000000..3ac1d89ad8
--- /dev/null
+++ b/crates/service/job/es-update-job/src/tasks/model_weights/util/copy_model_weight_records_to_documents.rs
@@ -0,0 +1,63 @@
+use std::time::Duration;
+
+use chrono::{DateTime, Utc};
+use elasticsearch::Elasticsearch;
+use sqlx::MySqlConnection;
+
+use errors::AnyhowResult;
+use mysql_queries::queries::model_weights::batch_get::batch_get_model_weights_for_elastic_search_backfill::batch_get_model_weights_for_elastic_search_backfill;
+use tokens::tokens::model_weights::ModelWeightToken;
+
+use crate::job_state::SleepConfigs;
+use crate::tasks::model_weights::util::create_model_weight_document_from_record::create_model_weight_document_from_record;
+
+pub struct Cursor {
+  pub token: ModelWeightToken,
+  pub updated_at: DateTime<Utc>,
+
+}
+pub async fn copy_model_weight_records_to_documents(
+  mut tokens: Vec<ModelWeightToken>,
+  mysql_connection: &mut MySqlConnection,
+  elasticsearch: &Elasticsearch,
+  sleep_config: &SleepConfigs,
+) -> AnyhowResult<Option<Cursor>> {
+  let mut last_updated_record = None;
+
+  while !tokens.is_empty() {
+    // NB: This list might be very large if we query from (1) the epoch, or (2) there was a large series of updates
+    let last = 50.min(tokens.len());
+    let drained_tokens = tokens.drain(0..last)
+        .into_iter()
+        .collect::<Vec<_>>();
+
+    let records =
+        batch_get_model_weights_for_elastic_search_backfill(
+          &mut *mysql_connection,
+          &drained_tokens
+        ).await?;
+
+    for record in records {
+      let new_cursor = Cursor {
+        token: record.token.clone(),
+        updated_at: record.updated_at.clone(),
+      };
+
+      create_model_weight_document_from_record(elasticsearch, record).await?;
+
+      let last_timestamp = last_updated_record.as_ref()
+          .map(|cursor: &Cursor| cursor.updated_at)
+          .unwrap_or_else(|| DateTime::UNIX_EPOCH);
+
+      if new_cursor.updated_at > last_timestamp {
+        last_updated_record = Some(new_cursor);
+      }
+
+      tokio::time::sleep(Duration::from_millis(sleep_config.between_es_writes_wait_millis)).await;
+    }
+
+    tokio::time::sleep(Duration::from_millis(sleep_config.between_job_batch_wait_millis)).await;
+  }
+
+  Ok(last_updated_record)
+}
diff --git a/crates/service/job/es-update-job/src/tasks/model_weights/util/create_model_weight_document_from_record.rs b/crates/service/job/es-update-job/src/tasks/model_weights/util/create_model_weight_document_from_record.rs
new file mode 100644
index 0000000000..a0f1466401
--- /dev/null
+++ b/crates/service/job/es-update-job/src/tasks/model_weights/util/create_model_weight_document_from_record.rs
@@ -0,0 +1,96 @@
+use anyhow::anyhow;
+use elasticsearch::{BulkOperation, BulkParts, Elasticsearch};
+use log::{error, info};
+use serde_json::Value;
+
+use elasticsearch_schema::documents::model_weight_document::{MODEL_WEIGHT_INDEX, ModelWeightDocument};
+use elasticsearch_schema::traits::document::Document;
+use enums::by_table::model_weights::weights_category::WeightsCategory;
+use errors::AnyhowResult;
+use mysql_queries::queries::model_weights::batch_get::batch_get_model_weights_for_elastic_search_backfill::ModelWeightForElasticsearchRecord;
+
+pub async fn create_model_weight_document_from_record(
+  elasticsearch: &Elasticsearch,
+  record: ModelWeightForElasticsearchRecord
+) -> AnyhowResult<()> {
+
+  info!("Create record for {:?} - {:?}", record.token, record.title);
+
+  let is_deleted = record.user_deleted_at.is_some() || record.mod_deleted_at.is_some();
+
+  let document = ModelWeightDocument {
+    token: record.token,
+
+    creator_set_visibility: record.creator_set_visibility,
+
+    weights_type: record.weights_type,
+    weights_category: record.weights_category,
+
+    title: record.title.to_string(),
+    title_as_keyword: record.title,
+
+    maybe_cover_image_media_file_token: record.maybe_cover_image_media_file_token,
+    maybe_cover_image_public_bucket_hash: record.maybe_cover_image_public_bucket_hash,
+    maybe_cover_image_public_bucket_prefix: record.maybe_cover_image_public_bucket_prefix,
+    maybe_cover_image_public_bucket_extension: record.maybe_cover_image_public_bucket_extension,
+
+    creator_user_token: record.creator_user_token,
+    creator_username: record.creator_username,
+    creator_display_name: record.creator_display_name,
+    creator_gravatar_hash: record.creator_gravatar_hash,
+
+    ratings_positive_count: record.maybe_ratings_positive_count.unwrap_or(0),
+    ratings_negative_count: record.maybe_ratings_negative_count.unwrap_or(0),
+    bookmark_count: record.maybe_bookmark_count.unwrap_or(0),
+
+    maybe_ietf_language_tag: match record.weights_category {
+      WeightsCategory::TextToSpeech => record.maybe_tts_ietf_language_tag,
+      WeightsCategory::VoiceConversion => record.maybe_voice_conversion_ietf_language_tag,
+      _ => None,
+    },
+
+    maybe_ietf_primary_language_subtag: match record.weights_category {
+      WeightsCategory::TextToSpeech => record.maybe_tts_ietf_primary_language_subtag,
+      WeightsCategory::VoiceConversion => record.maybe_voice_conversion_ietf_primary_language_subtag,
+      _ => None,
+    },
+
+    created_at: record.created_at,
+    updated_at: record.updated_at,
+    user_deleted_at: record.user_deleted_at,
+    mod_deleted_at: record.mod_deleted_at,
+
+    database_read_time: record.database_read_time,
+
+    is_deleted,
+  };
+
+  let op : BulkOperation<_> = BulkOperation::index(&document)
+      .id(document.get_document_id())
+      .into();
+
+  let response = elasticsearch
+      .bulk(BulkParts::Index(MODEL_WEIGHT_INDEX))
+      .body(vec![op])
+      .send()
+      .await?;
+
+  let json: Value = response.json().await?;
+
+  let had_errors = json["errors"].as_bool().unwrap_or(false);
+
+  if had_errors {
+    let failed: Vec<&Value> = json["items"]
+        .as_array()
+        .unwrap()
+        .iter()
+        .filter(|v| !v["error"].is_null())
+        .collect();
+
+    error!("Errors during indexing. Failures: {}", failed.len());
+
+    return Err(anyhow!("Errors during indexing. Failures: {}", failed.len()));
+  }
+
+  Ok(())
+}
diff --git a/crates/service/job/es-update-job/src/tasks/model_weights/util/mod.rs b/crates/service/job/es-update-job/src/tasks/model_weights/util/mod.rs
new file mode 100644
index 0000000000..6cb813b3fe
--- /dev/null
+++ b/crates/service/job/es-update-job/src/tasks/model_weights/util/mod.rs
@@ -0,0 +1,2 @@
+pub mod copy_model_weight_records_to_documents;
+pub mod create_model_weight_document_from_record;
\ No newline at end of file
diff --git a/crates/service/job/es-update-job/src/tasks/update_all_model_weights/mod.rs b/crates/service/job/es-update-job/src/tasks/update_all_model_weights/mod.rs
deleted file mode 100644
index 99408eaefc..0000000000
--- a/crates/service/job/es-update-job/src/tasks/update_all_model_weights/mod.rs
+++ /dev/null
@@ -1 +0,0 @@
-pub mod update_all_model_weights;
\ No newline at end of file
diff --git a/crates/service/job/es-update-job/src/tasks/update_all_model_weights/update_all_model_weights.rs b/crates/service/job/es-update-job/src/tasks/update_all_model_weights/update_all_model_weights.rs
deleted file mode 100644
index 373e8be48e..0000000000
--- a/crates/service/job/es-update-job/src/tasks/update_all_model_weights/update_all_model_weights.rs
+++ /dev/null
@@ -1,187 +0,0 @@
-use std::sync::Arc;
-use std::time::Duration;
-
-use anyhow::anyhow;
-use chrono::{DateTime, Utc};
-use elasticsearch::{BulkOperation, BulkParts, Elasticsearch};
-use log::{error, info};
-use serde_json::Value;
-
-use elasticsearch_schema::documents::model_weight_document::{MODEL_WEIGHT_INDEX, ModelWeightDocument};
-use elasticsearch_schema::traits::document::Document;
-use enums::by_table::model_weights::weights_category::WeightsCategory;
-use errors::AnyhowResult;
-use mysql_queries::queries::model_weights::batch_get::batch_get_model_weights_for_elastic_search_backfill::{batch_get_model_weights_for_elastic_search_backfill, ModelWeightForElasticsearchRecord};
-use mysql_queries::queries::model_weights::list::list_model_weight_tokens_updated_since::list_model_weight_tokens_updated_since;
-
-use crate::job_state::JobState;
-
-pub async fn update_all_model_weights(job_state: Arc<JobState>) {
-  // TODO(bt,2024-02-05): Write this cursor to Redis so job can resume without reindexing everything.
-  let mut cursor = DateTime::UNIX_EPOCH;
-
-  loop {
-    info!("Main loop; cursor @ {:?}", &cursor);
-
-    let result = with_database_main_loop(&mut cursor, &job_state).await;
-
-    if let Err(err) = result {
-      error!("Error in main loop: {:?}", err);
-      std::thread::sleep(Duration::from_millis(job_state.sleep_config.between_error_wait_millis));
-    }
-
-    // NB: Function should never return in success case, but we'll sleep just in case.
-  }
-}
-
-pub async fn with_database_main_loop(updated_at_cursor: &mut DateTime<Utc>, job_state: &JobState) -> AnyhowResult<()> {
-  info!("Acquiring MySQL connection...");
-
-  let mut mysql_connection = job_state.mysql_pool.acquire().await?;
-
-  let mut last_successful_update_at = *updated_at_cursor;
-
-  loop {
-    info!("Querying tokens updated since: {:?}", &updated_at_cursor);
-
-    let mut maybe_tokens =
-        list_model_weight_tokens_updated_since(&mut *mysql_connection, &updated_at_cursor).await?;
-
-    if maybe_tokens.is_empty() {
-      info!("No records updated since {:?}", &updated_at_cursor);
-      std::thread::sleep(Duration::from_millis(job_state.sleep_config.between_no_updates_wait_millis));
-      continue;
-    }
-
-    info!("Found {} updated records", maybe_tokens.len());
-
-    let mut last_observed_updated_at = *updated_at_cursor;
-
-    while !maybe_tokens.is_empty() {
-      // NB: This list might be very large if we query from (1) the epoch, or (2) there was a large series of updates
-      let last = 50.min(maybe_tokens.len());
-      let drained_tokens = maybe_tokens.drain(0..last)
-          .into_iter()
-          .map(|record| record.token)
-          .collect::<Vec<_>>();
-
-      let records
-          = batch_get_model_weights_for_elastic_search_backfill(&mut *mysql_connection, &drained_tokens).await?;
-
-      for record in records {
-        let updated_at = record.updated_at;
-
-        create_document_from_record(&job_state.elasticsearch, record).await?;
-
-        // NB: We don't want to advance the cursor to the current second just yet, because we might be processing
-        // several records with the exact same "updated_at" timestamp. We do know that due to ordering, we can update
-        // the timestamp cursor to the second before this timestamp, however.
-        //
-        // If we change the cursoring or batch sizes, we'll need to reconsider all of this logic.
-        if let Some(cursor_at_least) = updated_at.checked_sub_signed(chrono::Duration::seconds(1)) {
-          // NB: We don't want the cursor to slide backwards (it shouldn't, but the max will save us).
-          last_successful_update_at = cursor_at_least.max(last_successful_update_at);
-        }
-
-        *updated_at_cursor = last_successful_update_at.max(*updated_at_cursor);
-
-        last_observed_updated_at = updated_at;
-
-        std::thread::sleep(Duration::from_millis(job_state.sleep_config.between_es_writes_wait_millis));
-      }
-
-      std::thread::sleep(Duration::from_millis(job_state.sleep_config.between_job_batch_wait_millis));
-    }
-
-    // NB: The last cursor math put us a second behind the current clock. Here we'll advance to the last record.
-    // NB: This technically could miss records if we're updating within the same second batch we read from, but that
-    // seems both unlikely and not worth solving at our present scale.
-    *updated_at_cursor = last_observed_updated_at.max(*updated_at_cursor);
-
-    info!("Up to date at cursor = {:?}", &updated_at_cursor);
-
-    std::thread::sleep(Duration::from_millis(job_state.sleep_config.between_query_wait_millis));
-  }
-}
-
-async fn create_document_from_record(elasticsearch: &Elasticsearch, record: ModelWeightForElasticsearchRecord) -> AnyhowResult<()> {
-  info!("Create record for {:?} - {:?}", record.token, record.title);
-
-  let is_deleted = record.user_deleted_at.is_some() || record.mod_deleted_at.is_some();
-
-  let document = ModelWeightDocument {
-    token: record.token,
-
-    creator_set_visibility: record.creator_set_visibility,
-
-    weights_type: record.weights_type,
-    weights_category: record.weights_category,
-
-    title: record.title.to_string(),
-    title_as_keyword: record.title,
-
-    maybe_cover_image_media_file_token: record.maybe_cover_image_media_file_token,
-    maybe_cover_image_public_bucket_hash: record.maybe_cover_image_public_bucket_hash,
-    maybe_cover_image_public_bucket_prefix: record.maybe_cover_image_public_bucket_prefix,
-    maybe_cover_image_public_bucket_extension: record.maybe_cover_image_public_bucket_extension,
-
-    creator_user_token: record.creator_user_token,
-    creator_username: record.creator_username,
-    creator_display_name: record.creator_display_name,
-    creator_gravatar_hash: record.creator_gravatar_hash,
-
-    ratings_positive_count: record.maybe_ratings_positive_count.unwrap_or(0),
-    ratings_negative_count: record.maybe_ratings_negative_count.unwrap_or(0),
-    bookmark_count: record.maybe_bookmark_count.unwrap_or(0),
-
-    maybe_ietf_language_tag: match record.weights_category {
-      WeightsCategory::TextToSpeech => record.maybe_tts_ietf_language_tag,
-      WeightsCategory::VoiceConversion => record.maybe_voice_conversion_ietf_language_tag,
-      _ => None,
-    },
-
-    maybe_ietf_primary_language_subtag: match record.weights_category {
-      WeightsCategory::TextToSpeech => record.maybe_tts_ietf_primary_language_subtag,
-      WeightsCategory::VoiceConversion => record.maybe_voice_conversion_ietf_primary_language_subtag,
-      _ => None,
-    },
-
-    created_at: record.created_at,
-    updated_at: record.updated_at,
-    user_deleted_at: record.user_deleted_at,
-    mod_deleted_at: record.mod_deleted_at,
-
-    database_read_time: record.database_read_time,
-
-    is_deleted,
-  };
-
-  let op : BulkOperation<_> = BulkOperation::index(&document)
-      .id(document.get_document_id())
-      .into();
-
-  let response = elasticsearch
-      .bulk(BulkParts::Index(MODEL_WEIGHT_INDEX))
-      .body(vec![op])
-      .send()
-      .await?;
-
-  let json: Value = response.json().await?;
-
-  let had_errors = json["errors"].as_bool().unwrap_or(false);
-
-  if had_errors {
-    let failed: Vec<&Value> = json["items"]
-        .as_array()
-        .unwrap()
-        .iter()
-        .filter(|v| !v["error"].is_null())
-        .collect();
-
-    error!("Errors during indexing. Failures: {}", failed.len());
-
-    return Err(anyhow!("Errors during indexing. Failures: {}", failed.len()));
-  }
-
-  Ok(())
-}