Chinh (lelouvincx) lelouvincx

Note

• About estimating cost for releasing auto-generated metadata & commit.

• Context:

  • AI squad will release the features auto-generated dataset/data model metadata and auto-generated commits.
  • Offer these free (because it benefit our product).
  • The need to estimate the cost margin we have to pay OpenAI (I’m not sure about this metric, can be improved).
    • The cost will be pretty small (<= $5) to care about.
  • From data side, this task isn’t worth doing because too small ROI.
  • I suggest if we want to measure (and estimate) cost, let’s consider an LLM observability & tracing system.

• Done

  • DONE Check again the table Career Ladder

• DONE Outline the presentation and scripts

Note

• Holistics sharing - famous programmers share their development experience with AI (rảnh thì vô đọc): https://www.notion.so/holistics/Applying-AI-in-development-sharing-225f89dc7e49808e8f28cebc65b6811d

• Understands all basic chart types and when to use them.

• Based on carrier ladder, I can practice in sandbox mode to improve my skills.

• Do weekly practice:

  • “Business Question of the Week” exercise.
  • Business Translation Playbook 😼.

• Done

  • DONE Self review

• DONE Transfer test cases from existing car_retails testsuite

• DONE Review https://github.com/holistics/prefect/pull/342

• Note

  • The Speed Reading Book:
    • Mình nghĩ thế giới tương lai sẽ có phần giống trong Dune, khi người ta cấm robot và trí tuệ nhân tạo, đồng thời đề cao nâng cao năng lực của bộ não con người, như các thuật của Bene Gesserit.
    • Cuốn này mình đã định đọc từ lâu nhưng trì hoãn, đợt này tốt nghiệp rảnh rồi, mình sẽ đọc và thực hành.
    • TBU.

Done

• DONE Hand over tracking Break Down to Hieu

Note

• Một ảnh hưởng tích cực của các công cụ LLM Coding mà mình nhận ra là nó có thể giúp tăng tốc độ phát triển side projects/small startups. Code càng ngày càng rẻ hơn và chi phí cho việc plan/bảo trì/bug fixing mới là thứ đáng giá.

• Done

  • DONE Lên plan timeline cho Amya

• DONE Write review for a Triet

Note

• https://spiess.dev/blog/how-i-use-claude-code

  • One inline prompt to directly get the result: claude -p "your question".
    • Can create alias to auto-generated git commit message like: claude -p "Look at the staged git changes and create a summarizing git commit title following conventional commits. Only respond with the title and no affirmation.".
  • When I see something go wrong, press Escape immediately and edit the prompt.
  • Hidden command: /terminal-setup.
  • Can paste an image directly by copy paste.
  • Press escape twice and edit the previous prompt.

• Done

  • DONE Write review for Hieu and Dong

• DONE Find out how Zoho billing calculates 315 active customers

Note

• Tagging system:

  • Check lai trong code search còn track không.
  • Track at tenant-wise only.
  • Chia priority of tracked metrics.

• Supervisor Agent: Orchestrates workflow and escalates edge cases.

• Specialized Sub-agents: Sourcing, screening, engagement, compliance.

• Human-in-the-Loop Triggers: Strategic intervention points for quality control.

• Focus on augmenting human recruiters by automating top-of-funnel screening while ensuring viable candidates reach human review.

• Done

  • DONE Check SQL queries generated from the AI

• DONE Coi màn hình mua cho Linh

Note

• https://lucumr.pocoo.org/2025/7/3/tools/

  • Before thinking of automating some task with software/AI, ensure that the task works correctly as scale.

• Goals (the tests should be):

  • P1 Cover enough capabilities of our Explore Data AI.
    • From this, we have a sense of:
      • What are advantages.
      • What are disadvantages.
      • What are capabilities that our AI can outperform competitors'.

• Done

  • DONE Tagging system event tracking

• DONE Review PR for Hieu

• DONE Handle the schedule issue

• Done

  • DONE Push first 5 tests

• DONE Review the compiled SQL to see if any errors

• Note

  • Building an eval, which includes a dataset and a choice of evaluation methods.
  • Goal:
    • Structure a process of building evals for Holistics AI.
    • Should be composable with:
      • Multiple datasets.
      • Multiple evaluation methods.
      • Multiple system prompts.
      • Multiple LLMs.

• Done

  • DONE Check for anh Tien