Chinh (lelouvincx) / 2026-03-23

Created Mon, 23 Mar 2026 00:00:00 +0000 Modified Mon, 25 May 2026 06:02:25 +0000
784 Words
journal

Notes

  • Good timing does matter

    • Context: when I want to propose an idea to dbx team about setting up a Claude connector to interact with dbdiagram right inside claude desktop: Ampcode
    • I can read it from each team’s sync to quickly decide whether to propose an idea
    • For example, when I want to propose ((69c10475-619d-42df-b63e-1884b27ce548)), read the dbx team sync to find out whether they are having some similar key words in mind

  • dbt fails debugging process (collected from Thuan)

    • Check log in prefect, view error log
    • In model, there are 3 CTEs, run each cte step by step to find out where the error happen, check upstream
    • Find out the root cause
    • Proposing solutions

  • Surface interactive dbdiagram into claude/chatgpt (chat interface)

    • Context, from these:
    • Use case
      • When doing data support tickets, which are mostly data modeling and AQL writing problems, I’d like a quick way to reproduce: by imagining the dataset diagram from customers’ questions (as natural language) or AML models/datasets logseq.order-list-type:: number
        • Input: questions (as text) or AML models/datasets (they copy-paste into the ticket) logseq.order-list-type:: number
        • Output: an interactive interface of DBML diagram that lives inside the chat, can interact like logseq.order-list-type:: number
        • Currently I am doing it with mermaid logseq.order-list-type:: number
          • However, mermaid can not drag-and-drop tables, can not rearrange tables, which is important to me because: logseq.order-list-type:: number
            • I want to focus on some tables (fact tables) to design metrics on them. logseq.order-list-type:: number
            • Some diagram may contain hundreds of tables, which is hard to organize and focus on some tables logseq.order-list-type:: number
          • Mermaid’s syntax of relationship is harder to read (in my opinion). logseq.order-list-type:: number
            • logseq.order-list-type:: number
              erDiagram
    CUSTOMER ||--o{ ORDER : places
    ORDER ||--|{ LINE-ITEM : contains
    CUSTOMER {
        string name
        string email
        string address
    }
    ORDER {
        int orderNumber
        string deliveryAddress
    }
    LINE-ITEM {
        string productCode
        int quantity
        float pricePerUnit
    }
          • An exporting to dbdiagram feature so that teammates can collab on logseq.order-list-type:: number
        • If I use DBML instead: logseq.order-list-type:: number
          • I have to manually copy-paste DBML code from claude into dbdiagram to visualize and vice versa, just to have the visual view logseq.order-list-type:: number
      • I pop this idea out of my head because of this use case
    • Benefits
      • With an increasing number of users using AI, we can promote DBML more, pitch it as a powerful database designing and documenting language
      • New acquisition channel

  • Tasks

  • Partly today

    • DONE Revise phase 2 of MRR problem

    • Discussed with anh Hieu

      • After PR#812, these are statuses of problems:
        • 1.1 and 1.2 (static exchange rate in dbt)
        • 2.1 (missing holistics region when processing manual mapping logic, dbt): Fixed
        • 2.2 (domain mapping between tenant & company)
        • 2.3 (duplicated companies in hubspot - ICEYE case)
        • 2.4
        • 3.1
        • 3.2
        • 4
      • Next step (according to anh Hieu) is:

  • DONE Check around failed dbt tests (snowplow events) in #data-ops-bot

  • LEARNING The bigquery agent skill often get trouble in mapping logical (in dbt) and physical table names (in bigquery). How can I improve this?

    • Take dynamic mapping table from INFORMATION_SCHEMA (excluding dev_* if in prod)

  • DONE Write notion document

  • DONE Implementation

  • Done

    • DONE Teach Duc Anh

  • DONE Guide Thuan to solve the linear DAT-567

    • Talked and mentor him in person.
    • Next up I will interview to collect his thought process.
      • ((69c10ce7-08da-41dd-b384-a93312902245))

  • DONE Ping Anurag via DM

    • Had a discussion earlier with Anurag, action items:
      • Anurag will ingest the data of 6 countries from 6 RDS databases into Redshift.
        • Schema: airbyte_schema_* (sg, my, nw, etc).
        • For first step, only need to ingest Retailer tables.
      • Me fixes issue “history-only (589 retailers) from different countries”.
      • After that we’ll look closer into point 2 and 3.
    • Next action from me: waiting for Anurag to ingest data.

  • DONE UIUX learning points from SHUB

  • DONE Add playwright skills to understand web structure