danielrosehill commited on
Commit
2fd9fb7
·
verified ·
1 Parent(s): e92ffc9

Add 1 validation take

Browse files

Pushed by the capture app's Validation page.

data/validation-20261102T205354Z-enzy.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:66c1074054a2e4834cd6b7cd29877ecaa04ddc3edbaac707f5b11ed997b31819
3
+ size 4754413
meta/validation.jsonl CHANGED
@@ -11,3 +11,4 @@
11
  {"id":"20261102T205138Z_imn86i","take_id":"20261102T205138Z_imn86i","text":"Data team update. The nightly ETL job moved from cron scripts to Airflow, with one DAG per source system. Each task writes Parquet files to the data lake, and dbt builds the models on top in Snowflake.\n\nWe found duplicate rows in the orders table because the upstream API returns the same record on every page. We added a deduplication step on the primary key and a data quality test in dbt that fails the run if it happens again.\n\nThe dashboards moved from Tableau to Looker. Most of them load in under three seconds now, but the revenue dashboard still does a full join on two hundred million rows, so we are adding an aggregate table for it.\n\nOn the machine learning side, the churn model was retrained with XGBoost. AUC went from zero point seven eight to zero point eight three. The feature pipeline runs in Spark, and we log every experiment to MLflow.\n\nNext step is moving the model behind a FastAPI endpoint with Kubernetes autoscaling.","script_id":"tech-03","category":"tech","set_version":"v1.0","script_revision":"782485c52383","duration_s":56.5,"sample_rate":48000,"bit_depth":24,"device":"OnePlus Nord 3 5G · Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/156.0.0.0 Mobile Safari/537.36","microphone":"Yealink BH71","speaker":"Daniel Rosehill","created_at":"2026-11-02T20:51:38.008Z","shard":"data/validation-20261102T205145Z-l2uj.parquet","audio_path":"audio/tech-03/20261102T205138Z_imn86i.flac","audio_offset":38,"audio_length":5301496}
12
  {"id":"20261102T205247Z_ctvg8t","take_id":"20261102T205247Z_ctvg8t","text":"Mobile and frontend update. The Android app moved to Jetpack Compose for all new screens, and the iOS app is now fully on SwiftUI. Both apps share the business logic through Kotlin Multiplatform.\n\nCrash rate dropped below zero point one percent after we fixed a race condition in the image cache. Firebase Crashlytics was showing it on older Samsung phones with low memory.\n\nOn the web, we upgraded to Next.js fifteen with the app router and server components. Time to first byte went down by forty percent, and the Lighthouse score is now ninety-five on mobile.\n\nWe also added end-to-end tests with Playwright that run on every pull request, and visual regression tests with Storybook and Chromatic.\n\nAccessibility: every button now has an ARIA label, the color contrast passes WCAG AA, and the screen reader flow was tested with TalkBack and VoiceOver.\n\nRelease goes to the Play Store and the App Store on Thursday, with a staged rollout starting at ten percent.","script_id":"tech-04","category":"tech","set_version":"v1.0","script_revision":"f3e1db0630d1","duration_s":59.15,"sample_rate":48000,"bit_depth":24,"device":"OnePlus Nord 3 5G · Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/156.0.0.0 Mobile Safari/537.36","microphone":"Yealink BH71","speaker":"Daniel Rosehill","created_at":"2026-11-02T20:52:47.989Z","shard":"data/validation-20261102T205256Z-8lnv.parquet","audio_path":"audio/tech-04/20261102T205247Z_ctvg8t.flac","audio_offset":38,"audio_length":5527348}
13
  {"id":"20261102T205248Z_wkckbo","take_id":"20261102T205248Z_wkckbo","text":"Mobile and frontend update. The Android app moved to Jetpack Compose for all new screens, and the iOS app is now fully on SwiftUI. Both apps share the business logic through Kotlin Multiplatform.\n\nCrash rate dropped below zero point one percent after we fixed a race condition in the image cache. Firebase Crashlytics was showing it on older Samsung phones with low memory.\n\nOn the web, we upgraded to Next.js fifteen with the app router and server components. Time to first byte went down by forty percent, and the Lighthouse score is now ninety-five on mobile.\n\nWe also added end-to-end tests with Playwright that run on every pull request, and visual regression tests with Storybook and Chromatic.\n\nAccessibility: every button now has an ARIA label, the color contrast passes WCAG AA, and the screen reader flow was tested with TalkBack and VoiceOver.\n\nRelease goes to the Play Store and the App Store on Thursday, with a staged rollout starting at ten percent.","script_id":"tech-04","category":"tech","set_version":"v1.0","script_revision":"f3e1db0630d1","duration_s":59.15,"sample_rate":48000,"bit_depth":24,"device":"OnePlus Nord 3 5G · Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/156.0.0.0 Mobile Safari/537.36","microphone":"Yealink BH71","speaker":"Daniel Rosehill","created_at":"2026-11-02T20:52:48.473Z","shard":"data/validation-20261102T205256Z-8lnv.parquet","audio_path":"audio/tech-04/20261102T205248Z_wkckbo.flac","audio_offset":5527420,"audio_length":5527348}
 
 
11
  {"id":"20261102T205138Z_imn86i","take_id":"20261102T205138Z_imn86i","text":"Data team update. The nightly ETL job moved from cron scripts to Airflow, with one DAG per source system. Each task writes Parquet files to the data lake, and dbt builds the models on top in Snowflake.\n\nWe found duplicate rows in the orders table because the upstream API returns the same record on every page. We added a deduplication step on the primary key and a data quality test in dbt that fails the run if it happens again.\n\nThe dashboards moved from Tableau to Looker. Most of them load in under three seconds now, but the revenue dashboard still does a full join on two hundred million rows, so we are adding an aggregate table for it.\n\nOn the machine learning side, the churn model was retrained with XGBoost. AUC went from zero point seven eight to zero point eight three. The feature pipeline runs in Spark, and we log every experiment to MLflow.\n\nNext step is moving the model behind a FastAPI endpoint with Kubernetes autoscaling.","script_id":"tech-03","category":"tech","set_version":"v1.0","script_revision":"782485c52383","duration_s":56.5,"sample_rate":48000,"bit_depth":24,"device":"OnePlus Nord 3 5G · Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/156.0.0.0 Mobile Safari/537.36","microphone":"Yealink BH71","speaker":"Daniel Rosehill","created_at":"2026-11-02T20:51:38.008Z","shard":"data/validation-20261102T205145Z-l2uj.parquet","audio_path":"audio/tech-03/20261102T205138Z_imn86i.flac","audio_offset":38,"audio_length":5301496}
12
  {"id":"20261102T205247Z_ctvg8t","take_id":"20261102T205247Z_ctvg8t","text":"Mobile and frontend update. The Android app moved to Jetpack Compose for all new screens, and the iOS app is now fully on SwiftUI. Both apps share the business logic through Kotlin Multiplatform.\n\nCrash rate dropped below zero point one percent after we fixed a race condition in the image cache. Firebase Crashlytics was showing it on older Samsung phones with low memory.\n\nOn the web, we upgraded to Next.js fifteen with the app router and server components. Time to first byte went down by forty percent, and the Lighthouse score is now ninety-five on mobile.\n\nWe also added end-to-end tests with Playwright that run on every pull request, and visual regression tests with Storybook and Chromatic.\n\nAccessibility: every button now has an ARIA label, the color contrast passes WCAG AA, and the screen reader flow was tested with TalkBack and VoiceOver.\n\nRelease goes to the Play Store and the App Store on Thursday, with a staged rollout starting at ten percent.","script_id":"tech-04","category":"tech","set_version":"v1.0","script_revision":"f3e1db0630d1","duration_s":59.15,"sample_rate":48000,"bit_depth":24,"device":"OnePlus Nord 3 5G · Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/156.0.0.0 Mobile Safari/537.36","microphone":"Yealink BH71","speaker":"Daniel Rosehill","created_at":"2026-11-02T20:52:47.989Z","shard":"data/validation-20261102T205256Z-8lnv.parquet","audio_path":"audio/tech-04/20261102T205247Z_ctvg8t.flac","audio_offset":38,"audio_length":5527348}
13
  {"id":"20261102T205248Z_wkckbo","take_id":"20261102T205248Z_wkckbo","text":"Mobile and frontend update. The Android app moved to Jetpack Compose for all new screens, and the iOS app is now fully on SwiftUI. Both apps share the business logic through Kotlin Multiplatform.\n\nCrash rate dropped below zero point one percent after we fixed a race condition in the image cache. Firebase Crashlytics was showing it on older Samsung phones with low memory.\n\nOn the web, we upgraded to Next.js fifteen with the app router and server components. Time to first byte went down by forty percent, and the Lighthouse score is now ninety-five on mobile.\n\nWe also added end-to-end tests with Playwright that run on every pull request, and visual regression tests with Storybook and Chromatic.\n\nAccessibility: every button now has an ARIA label, the color contrast passes WCAG AA, and the screen reader flow was tested with TalkBack and VoiceOver.\n\nRelease goes to the Play Store and the App Store on Thursday, with a staged rollout starting at ten percent.","script_id":"tech-04","category":"tech","set_version":"v1.0","script_revision":"f3e1db0630d1","duration_s":59.15,"sample_rate":48000,"bit_depth":24,"device":"OnePlus Nord 3 5G · Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/156.0.0.0 Mobile Safari/537.36","microphone":"Yealink BH71","speaker":"Daniel Rosehill","created_at":"2026-11-02T20:52:48.473Z","shard":"data/validation-20261102T205256Z-8lnv.parquet","audio_path":"audio/tech-04/20261102T205248Z_wkckbo.flac","audio_offset":5527420,"audio_length":5527348}
14
+ {"id":"20261102T205349Z_0e5e8l","take_id":"20261102T205349Z_0e5e8l","text":"Quick update before Erev Shabbat, because this week was a proper balagan.\n\nThe arnona website kept throwing a 500 error, and the API was returning malformed JSON, davka on the day the tashlumim were due. I paid at the kupa and got a heshbonit mas kabala.\n\nThen Bituach Leumi wanted an ishur from the kupat cholim. I logged into the Maccabi app with OAuth, downloaded the PDF, and the upload timed out twice. Third time, beseder.\n\nAt work, tachles: a pod in the Kubernetes cluster had a memory leak, so I added a limit in the Helm chart and a Grafana alert. I also exported the speech model to ONNX, and latency on the CPU is sababa.\n\nAs an osek patur, I wrote a script that pulls invoices from Gmail into a CSV for Mas Hachnasa.\n\nThe mazgan is leaking, and the vaad bayit says it's the kablan's problem. Yalla. Shabbat shalom.","script_id":"mixed-01","category":"mixed","set_version":"v1.0","script_revision":"87063251fee2","duration_s":50.83,"sample_rate":48000,"bit_depth":24,"device":"OnePlus Nord 3 5G · Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/156.0.0.0 Mobile Safari/537.36","microphone":"Yealink BH71","speaker":"Daniel Rosehill","created_at":"2026-11-02T20:53:49.131Z","shard":"data/validation-20261102T205354Z-enzy.parquet","audio_path":"audio/mixed-01/20261102T205349Z_0e5e8l.flac","audio_offset":38,"audio_length":4750442}