Loop Engineering Practical Guide: How US Stock Investors Turn AI from Chatbot into Research Assistant

In the past two years, many people’s approach to using AI has essentially been a “Q&A” format.

Write a prompt, wait for a reply, check the result, add more prompts. AI seems to be working, but you’re still constantly directing, correcting, verifying sources, and deciding when to stop.

What really holds us back is that the workflow still requires human oversight: directing, correcting, verifying sources, and deciding when to stop.

Loop Engineering is designed to solve this problem.

It breaks repetitive work into a runnable loop: first define the goal and sources, then specify execution rules, validation criteria, retry on failure, state memory, and stopping conditions.

Prompt Engineering focuses on: how to ask AI. Loop Engineering focuses on: how to make AI consistently do a task to an acceptable standard.

Why Is Loop Engineering Suitable for US Stock Investors?

US stock investing is naturally suited for Loops.

Because every day you need to repeatedly look at similar information:

Indices, VIX, technical indicators, AI news, semiconductors, memory/storage sector, earnings, macro data, company announcements.

This information is not purely subjective; most of it is verifiable, and it’s worth tracking continuously.

So you can let it generate a verified research briefing every day based on a fixed process, and remember the market clues from the previous session.

That’s the transformation from a “chat tool” into a “research assistant.”

What Does a US Stock Research Loop Need?

Simply put, it needs six things:

• Heartbeat: When to run, e.g., Sydney time 8-9 AM daily.
• Work Definition: Which indices, sectors, companies, and news to look at.
• Skill: Fixed rules, such as data checking, news checking, writing style.
• Reviewer: Independently checks data, links, and conclusions.
• Connector: Can read/write files, fetch data, generate reports.
• State: Uses LOOP-STATE.md to record last key points and next focus areas.

The most important here are Reviewer and State.

The Reviewer prevents AI from confidently writing errors. The State prevents AI from starting from scratch every day.

What US Stock Tasks Are Suitable for a Loop?

Tasks suitable for a Loop generally meet three conditions:

Repetitive, verifiable, and worth automating.

For example:

• Daily US stock market research
• Weekly investment review
• AI news monitoring
• Semiconductor sector tracking
• Memory cycle tracking
• Earnings season company tracking
• Investment notes organization
• X content draft generation
• Watchlist updates

Common characteristics of these tasks: information-intensive, repeated, need verification, and long-term accumulation is valuable.

My First Practical Project: Daily US Stock Market Research Loop

My top priority is to implement a “Daily US Stock Market Research Loop.”

The goal is clear:

Every day at Sydney time 8-9 AM, generate a concise morning US stock market research report.

Content includes:

• Major US indices: S&P 500, Nasdaq, VIX
• Technical indicators: RSI if possible
• Semiconductor index: e.g., SOX or SMH
• Memory/storage sector: e.g., MU, STX, WDC
• AI news: 2-3 latest important news items
• Investment perspective commentary
• Action prompts
• X thread draft

But the truly important part is not “getting AI to write a report.”

The truly important part is: the data, links, news, and judgments in the report must be verified.

Below is the main prompt I will use directly.

Daily US Stock Market Loop Main Prompt

You are running in loop mode. This is your charter.

GOAL
Generate a concise morning US stock market research report daily, defaulting to Sydney time, run at Sydney time 8-9 AM.

The report must include:

1. Major US indices
   - S&P 500
   - Nasdaq
   - VIX
   - If available, include RSI or other key technical indicators

2. Semiconductor & AI Infrastructure Sector
   - SOX or SMH
   - Focus on NVDA, AMD, AVGO, TSM, ASML, etc.
   - Focus on AI data centers, inference chips, cloud capex trends

3. Memory/Storage Sector
   - MU, STX, WDC, etc.
   - Focus on HBM, AI server demand, HAMR, memory cycle changes

4. Latest AI News
   - Select 2-3 latest and important AI news items
   - Each must include a reliable source
   - Each must include an investment angle, not just a news summary

5. Overall Interpretation
   - What is the most important change in the market today?
   - Which signals are worth continuing to watch?
   - Which are just noise?
   - What insights for US stock investors?

6. Action Prompts
   - Can include a watchlist
   - Can include risk reminders
   - Can combine Peter Lynch-style fundamental thinking
   - Can include LEAPs, put strategies and other long-term investment tools
   - Do not write uncertain judgments as definitive investment advice

OUTPUT
Generate two deliverables:
1. Markdown version of the morning market research report
2. X thread draft

WHERE THE WORK IS
- Prioritize reliable sources: Yahoo Finance, TradingView, Reuters, company announcements, exchange data, authoritative financial media, etc.
- Read and update LOOP-STATE.md
- If a piece of data cannot be verified, it must be explicitly marked as "not fully verified" – do not pretend certainty.

HOW TO WORK
- Cross-verify at least 2 reliable sources
- All key data must include sources
- All news links must be accessible and content must match the summary
- Interpretation should be data-driven, practical, and include an Australian investor perspective
- All times default to Sydney time
- Anything that requires my decision: uniformly mark as "[requires me]"

HOW TO CHECK YOURSELF
Use reviewer-skill to independently audit:
1. Data accuracy
2. Link validity
3. News authenticity and currency
4. Style consistency
5. Investment judgment over-certainty
6. Alignment with Australian investor perspective and Sydney time

Retry each item a maximum of 3 times.
If still unverifiable after retries, retain the item but mark it "not fully verified."

HOW TO REMEMBER
Update LOOP-STATE.md with:
1. Summary of this report
2. Key market clues from this session
3. Companies/sectors worth continuing to watch
4. Data source issues
5. Topics to keep tracking next time

WHEN TO STOP
Stop when ALL of the following conditions are met:
1. Markdown report completed
2. X thread draft completed
3. Reviewer check passed
4. LOOP-STATE.md updated
5. All unverified content explicitly marked

Now begin: First read LOOP-STATE.md, then execute.

Reviewer Skill

The Reviewer’s role is auditing.

It should act like a read-only inspector, only returning PASS or FAIL. Checks include: whether index, VIX, RSI, semiconductor, and storage data are accurate; whether links work; whether AI news is authentic; whether unverified opinions are written as facts; whether uncertain judgments are written as definitive investment advice.

---
name: market-data-reviewer
description: Strictly read-only Reviewer, checks market data, semiconductor/storage, AI news accuracy and links.
---

You are an extremely rigorous Reviewer. You verify only, do not modify.

Your task is to check whether the daily US stock market research report is reliable.

Checklist:

1. All indices, RSI, VIX, semiconductor, storage data are accurate and sourced
2. All links work and content matches
3. AI news is reliable, current, and hallucination-free
4. Style aligns with data-driven + Australian investor perspective
5. All times default to Sydney time
6. Investment judgments are not disguised as deterministic predictions
7. Unverified opinions are not written as facts
8. Output includes Markdown report and X thread draft

Output only in the following format:

****PASS****: brief explanation why it passed

or

****FAIL****: list of specific issues

The Role of LOOP-STATE.md

LOOP-STATE.md is the memory core of this system.

It can record:

• User preferences
• Key sectors
• Reliable sources
• Summary of last report
• Current market clues
• Data source caveats
• Topics to continue tracking next time

For example, if today you notice unusual movement in the storage sector, tomorrow the AI should not start from scratch understanding it; this week you focus on HBM and AI data centers, next week’s review should know whether this line continues to be relevant.

Without a state file, AI “loses memory” every day.

With a state file, it begins to act like a continuous collaborative US stock research assistant.

Weekly Investment Review Loop

Once the daily loop stabilizes, you can expand into a weekly review loop.

The weekly version doesn’t just sum up daily content; it answers more critical questions:

• What was the real market theme this week?
• Which judgments were validated?
• Which judgments need correction?
• Which sectors are starting to strengthen?
• What risks are accumulating?
• What should be the focus next week?

The weekly loop can output a full weekly report, a long X thread draft, a next-week watchlist, and continue updating LOOP-STATE.md.

Risk Control

The biggest risk of a Loop is amplifying errors automatically.

Therefore, you must add several brakes:

• Clear stopping conditions
• Retry limits
• Key data must be verified
• Publishing, order placement, and other critical actions require manual confirmation
• Investment judgments cannot be written as deterministic predictions

The core of a Loop is not “auto-write,” but “auto-execute + auto-check + retain human decision-making.”

Conclusion

Loop Engineering provides clear value to US stock investors:

It transforms the daily repetitive tasks of information gathering, data checking, market review, and content drafting into a verifiable workflow.

You define the standards. AI executes the loop. The Reviewer checks quality. The State retains memory.

Finally, you only need to review the results, identify key points, and make critical decisions.

That’s how you move AI from a chat window into your personal US stock research system.