Chinh (lelouvincx) lelouvincx

Usually in data modeling-related tickets like this, to reproduce, we require:

• The how-to question.
• The dataset schema (which model, relationships, available dimensions/metrics).
• The expected AQL explore.

Usually in debugging a dbt test warning/failure:

• The error / signal.
• Which model? What are upstream models (to know the root cause can come from) and downstream models (to know affected models)?
• Query real data.
• Find out root cause. This may vary not only one root cause but many.
• Categorize root causes:
  • Data pipeline side
  • Source side
• Some common questions:
  • The test is testing what behavior? What does the test expect and not expect?
  • If we don’t feel the test make sense, question why is it there from first day?

In [[Oxygen Not Included]], there is a DLC that introduces a new kind of labor: Bionic Dupes.

[[1on1 Report - 2026-03-18]] notes

• Anh Triet’s concerns
  • AI (my primary focus on this)
    • How to integrate AI into development experience in data team?
    • What is required to make the team more productive?
  • Current data infra / pipeline being too complex
    • Anh Triet cant trust
    • Feeling it is being coupled into each other too much
  • PR Workflow (anh Triet feels no one is updating)

Recent thoughts about AI

• AI can make us worse but also can empower us

Tasks

Partly today

• TODO Think about applying AI into resolving #data-ops-bot issues (data pipelines)

Tasks

Partly today

• ((69a2a1cc-fc1f-4802-ba56-48183fa9e9a5))

Done

• DONE Showbie use case - embedded dashboard QA

• Requirements

  • Showbie’s product team wants to review and approve dashboards before they go live to end customers.
  • The PM needs to preview what the dashboard will look like when embedded in Showbie’s app (and not just an internal Holistics view that might look different).
  • Scope: both internal (within Holistics) and external (in Showbie’s app)
  • Where customers interact with dashboards.
  • Dashboards should not be visible to customers until they’ve been approved.
  • The QA process is specifically about verifying the visual design and presentation of dashboards, not just data correctness.

• Possible approaches:

https://vinted.engineering//2026/03/11/risk-based-testing/

• Pretty cool tip on tagging dbt models with high/low risk, high/low frequency and run the tests based on these tags.

Tasks

Partly today

• ((69b3d030-98dd-450d-88d8-e00c1c322c59))
• ((69a2a1cc-fc1f-4802-ba56-48183fa9e9a5))

Done

• DONE Debrief the last call with Showbie

((69af7837-e2f7-4a64-8ff0-d55a131c55ac))

DONE Debrief the last call with Showbie

• LEARNING Try out the Calculation Builder feature and prepare some use case I can use immediately in onboarding calls.
• LEARNING I feel like the calls I’m leading is not at my expected quality, I should practice more.
  • One of the way I’m thinking about this using past calls led by Huy/Vincent, immitate them: whenever a prospect asks for something, I pause and answer first, then play the video to see how Huy/Vincent answers => take learning points.

DONE Version v0.2.4

Consulting this week

• For too many issues, batch the truly small ones (DAT-219, DAT-268) into a single focused session, and explicitly defer or close ones that are no longer relevant. Carrying 15 open issues creates cognitive load.
• The root fix is sleep. Consider a hard cutoff time and protect 7+ hours.
• Watch the smartclass project does not become a rabbit hole. => Prefer setting a clear boundary for v0.1 and park it.
• This week the infrastructure work was necessary and well-executed, but the number of context switches is high. Consider blocking 2–3 “focus days” next week for the Census migration and Linear backlog triage.
  • I dont know whether it affects me, but I will keep aware of.

Claude Skill Building Guide

Just received performance review result for 2025H2. Very excited.

• Salary increase and year bonus: https://media.secondbrain.lelouvincx.com/2026/03/8af03c4fafbddbbd552e7850e668d81a.png
• Explore opencode and feedback it’s pretty good.
  • Especially models GLM-4.7.
  • Even claude opus/sonnet there is much cheaper than ampcode.
  • This can replace ampcode in some small, repetitive, context gathering tasks.
  • Problem: lacking of an easy way to read/share threads like ampcode.

Tasks

• ((69b21fb9-6f3d-4c38-8483-892aa9d89cac))

Done

Tasks

Done

• TODO Search for a tool to summarize youtube videos

TODO Check failed dbt tests in #data-ops-bot, list all issues, then create corresponding linear issues for them

DONE DAT-554

DONE DAT-555

fct_job_queue_performance

• 1-Ampcode 2-Ampcode 3-Ampcode (fix) 4-Ampcode (migrate) 5-Ampcode (migrate continue)
• PR dbt (reviewing): https://github.com/holistics/dbt/pull/854
• PR Holistics (draft): https://github.com/holistics/internal-aml-project/pull/69
• This model is used by dataset Job Queue Performance, then used by dashboard [WIP] Job Queue Performace 3.0
  • Not a big impact because no one is using this dashboard right now.
  • The team is probably using this dashboard Report Job Queue Performance Monitoring.
• But is this dbt model essential? Worth keeping?
  • This model estimates, at minute-level, how loaded each tenant queue is and how many were waiting vs total active jobs.
    • Minute-level => very big.
  • https://holistics.slack.com/archives/C79QB3FEV/p1755662645561689
    • There is already a solution for this: dashboards those let customers monitor directly in their tenant (via embedding). Similar to Usage Monitoring.
    • In Job Monitoring -> Analytics.
  • This dbt model is staled.
• Should we remove it?
  • Is there any use case in the future that we can consider using it?
    • I don’t think so. The only use case is Job Monitoring, which is already addressed by squad performance.
  • Is it safe if we remove this model? Yes, we can recreate it from fct_jobs because that model already has key inputs:
    • created_at, start_time, end_time
    • assigned_queue
    • queue_limit
    • queue_valid_from
    • tenant/region keys
  • The cost for storing this table: <$1/month.
  • Yes. Remove it.
• DONE Write a notion document about this and let teammate review => https://www.notion.so/holistics/DE-212-336f89dc7e498006a2a0e78622de854c

DONE Write downstream models

99

• {{video https://www.youtube.com/watch?v=ws9zR-UzwTE}}
• Impressions
  • He pitched it good. There’s gap between people who do hand programming and people who vibe code 100%. He still wants to use AI, but does not want to give 100% permission on coding agents like how claudecode does: grep and text edit all the time.
  • https://media.secondbrain.lelouvincx.com/2026/03/442d453b9c1e9273d432c8d302e921bc.png

Listing pending issues from linear into logseq

• DAT-450 Census → HubSpot reverse ETL issue causing stale customer status (e.g., Telio) due to rejected updates.
• DAT-469 Call-booking data appears to stop after Nov 2024 in lead conversion analytics; needs investigation.
• DAT-510 Validate holistics-customers logic and add a dashboard tab for period-based customer movement metrics.
• DAT-524 Exclude internal Zoho test accounts from churn metrics by fixing filters in Zoho events pipeline.
• DAT-155 Investigate why active users have censored emails in BigQuery customer dataset/report.
• DAT-163 Improve Prefect production secret management (currently manual), likely toward a vault/as-code workflow.
• DAT-219 Migrate deprecated Prefect String usage to Variable.
• DAT-259 Ensure dim_sales_reps keeps historical/ex-rep records instead of nulling old emails.
• DAT-268 Use Python tempfile in Prefect flows to avoid persistent temp files and cleanup issues.
• DAT-283 Expand Calendly ingestion to all call types (not only Demo/Onboarding).
• DAT-371 Investigate Census/HubSpot naming/status inconsistency around MoneySmart/Parts records.
• DAT-462 JSONB→STRING conversion in BigQuery breaks JSON parsing and impacts downstream logic/reports.
• DAT-463 Follow-up to identify other places impacted by the same root issue.
• DAT-504 Investigate mismatch between user list and active-user count in Pencil tenant health dashboard.
• DOC-36 Merge and simplify cross-model calculation docs into one clearer official guide.
• I will re-prioritize and complete them as soon as possible. They are all small tasks.

Tasks

Done

• TODO Think about how would a normal data team adopt a BI chatbot

Tasks

Done

• DONE Plan building a live demo dashboard for this slide

• https://media.secondbrain.lelouvincx.com/2026/03/ae681d144e135a4a657287680de4a261.png

DONE Check Prefect slack automation failure

Not sure the exact problem, I guess it’s because of version conflict.

DONE Simplify pyproject.toml and deploy to check it first.

Run flows to debug.

The root cause is when upgrading packages using dependabot, we accidentally upgrade a small package used by prefect to post notification to slack. Details in the PR.

IDEA Do “performance review” my self by every 3 month.

• Next target is end of this month.

Intalled opencode today, using model GPT-5.3 Codex for free. Will leave reviews here.

Tasks

Done

• ((69a2a1cc-fc1f-4802-ba56-48183fa9e9a5))

• ((69af7837-e2f7-4a64-8ff0-d55a131c55ac))

• ((69a81acd-1ec9-457d-ab25-abb87c57d581))

• DONE Smartclass v0.1 (to be updated)

• Project setup: wrangler.toml, D1/R2 bindings, Hono worker entry.

• D1 schema migrations (users, exercises, answer_schemas, submissions, lectures).

• Auth: phone+password login/register, JWT middleware, teacher creates students (pw 123), pending approval flow.