Video summary

The Simple 4-Step Process To Build Your Own AI Trading Assistant With Claude (for Beginners)

Main summary

Key takeaways

Technology

Summary of the Video

The video demonstrates a beginner-friendly 4-step workflow for building an AI trading assistant using Claude Code—an agentic AI that can edit/build files, not just chat. The creator claims the full trading dashboard/assistant can be built in a single afternoon by repeating the same workflow.

The main takeaway is that the workflow is repeatable, not that the dashboard must match the exact design shown.

What the Assistant Does (Example End Result)

The example assistant produces a personal HTML “TraderView-like” trading dashboard: a private single-page website that:

  • Tracks trades and updates constantly based on user inputs.
  • Flags repeat mistakes by identifying tendencies/patterns from trade write-ups/journals.
  • Auto-fills the dashboard from daily inputs (the user provides information via prompts).
  • Runs a scheduled daily “ritual” at about 4:15 PM using Claude Code routines to ask for:
    • daily trade journal info

The 4-Step Workflow (Core Tutorial)

1. Plan Mode (No Edits Yet)

In this step, Claude Code is configured to not modify files. The goal is to generate an implementation plan as a .md file.

Key technique: use an “ask user question” / “ask me questions” mode so Claude collects requirements through multiple-choice/dropdown-style questions (more token-efficient than open-ended chat).

The plan prompt includes:

  • Desired output: a local HTML single-page app opened in a browser.
  • Dashboard structure: tabs, stats, and features such as:
    • trade tags/patterns
    • “best ops” (meaning explained in the video)
    • a coach/accountability view
  • Data handling preferences:
    • either auto-pull trades via a script, or
    • manual entry (the creator’s example uses manual + daily routine)
  • References/templates:
    • screenshots/templates (TraderView-like or TradingView-inspired layouts) so Claude has a clear visual target.

2. Build Mode (Execute the Plan)

Paste the generated .md plan back into Claude Code, but switch from plan mode to editing enabled.

Claude then builds the skeleton of the dashboard:

  • produces a working HTML website
  • likely provides a preview link

3. Personalize (Iterative Tweaks)

Once the skeleton exists, Claude Code is used to apply small changes iteratively, such as:

  • adding/modifying stats and chart axes (e.g., exact accuracy %, cumulative P&L on the X axis)

  • adding an additional “tendencies/coach” layer

The “coach” component analyzes trade write-ups to detect:

  • tendencies/mistakes, such as repeated “no man’s land sizing” when the user’s text mentions it
  • patterns it can recognize itself versus patterns the user explicitly instructs it to watch

The creator describes this stage as the “fun” part—where the assistant becomes tailored to the user’s personal grading and review habits.

4. Make It Repeatable (Routines / Scheduling)

Use Claude Code routines to schedule the daily workflow (e.g., 4:15–4:30 PM).

The routine prompts the user with questions like:

  • “What trades did you take today?”
  • “Add anything to the journal / best ops”

Purpose: convert the dashboard into a consistent daily DRC (daily review checklist/ritual) tool, not a one-time build.

Specific Guidance & Claims About Efficiency/Quality

  • Token efficiency tip: Using ask-user-question mode reduces tokens versus free-form answers.
  • Architecture-first approach: Planning first reduces debugging because edits happen only after the plan is complete.
  • Templates/screenshot examples help Claude match desired layout and UI structure.
  • Input quality matters: Claude’s usefulness depends heavily on the quality of inputs (trade write-ups, playbooks, grading details). Compute alone won’t guarantee “smart” outputs.
  • Caution about over-automation: Don’t let the AI make decisions. Use it for insight, review, and pattern detection, while the user stays in control.

Features Highlighted in the Example Dashboard

  • Trade history + grading-based views (split by grades like A+, A, A-, etc.)
  • A Tendencies page that groups repeated patterns found across multi-year write-ups
  • A coach note generated after the user enters daily trades/journal content, including:
    • mistakes made that day (e.g., “two no man’s land sizing today”)
  • A playbook-focused section that includes:
    • win rate / P&L by pattern
    • “checks in favor” and “rules” describing how trades should be executed

Product/Strategy Alternatives Discussed

  • The assistant is positioned as an alternative/improvement to tools like TraderView:
    • a custom version with personal grading accuracy
    • less convoluted tagging
  • A future idea: a dynamic resource page based on the user’s “businesses,” catalysts, setups, and “checks in favor,” potentially including guidance on grading setups.

Main Speakers / Sources

  • Tim (main demonstrator) Builds the Claude Code trading dashboard workflow and explains the tendencies/coach layer.

  • Garrett (co-host/interviewer) Asks questions about workflow steps and use cases.

  • “Trading Floor podcast” (mentioned as show context) Includes promotional mentions for smbtradingfloor.com.

Original video