A $6–8 Million Hedge Fund Research Team, I Built It Into an App Using Prompts#

I built an AI Mini App using prompts: You can have one too, even without coding skills.

Part 1 of "A One-Person AI Hedge Fund":

First, let's get "Institution-Grade Polished Research Reports" working (Gemini Beginner Edition)

A bit of background first: I used to be a researcher at a top-10 AUM hedge fund in Europe, and before that, a data scientist at a major US internet company. Whether it's a hedge fund-level research system or building trading strategies based on massive datasets, today both can be replicated with "Cloud Services + AI" at a much lower barrier to entry. I'll share my complete journey from 0 to 1, and together we can build this system stronger.

This has made me even more certain of one thing: The biggest difference between retail investors and hedge funds is often not "who is smarter or has more accurate views," but whether there is a set of organizational processes that can operate continuously—division of labor, standardization, evidence, arbitration, and delivery. The most expensive part of fund research has never been the views themselves, but the organization.

First, See the Result: I just input a news headline, and it spits out a research memo from a fund's Chief Investment Officer#

Input: A headline from a Coindesk news article dated 20260301

Output: A complete research report generated by a hedge fund research desk 👇

Don't rush to learn how to build it yet. First, confirm: Is this deliverable worth you saving?

Why is a "Research Desk" so expensive?#

Because the money is almost entirely burned on people and data, specifically "organized human resources."

The annual run-rate for a small multi-asset research desk (research only, excluding trading systems/custody/fund administration) is roughly $6–8 million. The structure typically looks like this:

Human Capital Costs: The majority, around 80% ($5–6M)

Your desk essentially needs 8–10 core intellectual positions: Research PM/Research Secretary (Atlas), Macro, Sectors, Single Stocks, Crypto, Derivatives & On-chain, Risk, Quantitative Validation, plus a CIO for final arbitration.

These roles at a hedge fund are priced not on "writing reports," but on being accountable for results. Therefore, compensation = base salary + bonus + benefits + employment costs. Stacked up over a year, it reaches the multi-million dollar level.

Data & Research Tools: Around 20% ($0.2–1.5M)

Terminals (Bloomberg/Refinitiv/FactSet, etc.), news/research reports, market/macro data, on-chain/derivatives data, and various internal data cleaning and subscription services. A small team buying lightly is tens of thousands of dollars; buying heavily can exceed a million.

The essence of the cost is not the information itself, but "organizational capability"—what you're paying for is:

Division of Labor (Each person only does their part, depth emerges)
Parallel Processing (Multiple tracks advance simultaneously, speed emerges)
Standardization (Unified templates enable review, iteration, handover)
Evidence (Every judgment can be verified by data anchors)
Arbitration (The CIO compresses disparate views into "language that can be traded")

We can write this "organizational structure" into a prompt workflow, allowing one person to have the same deliverables.

"A One-Person AI Hedge Fund": This is Just Part 1#

What I want to do with this series is: Upgrade personal investing from "based on feeling" to a "reusable, iterable, automatically running" system.

This part only solves one thing: Beginners with no coding skills can quickly produce a polished, shareable research deliverable.
I will gradually upgrade it to:
- Data scraping and calculation verification (scripted, reproducible)
- Automated operation (scheduled runs, push notifications, review)
- Packaging it into a skill, connecting to OpenClaw / agent runtime to let it run on its own

Don't think too far ahead today: First, produce the "deliverable." Once the deliverable exists, the system has a foothold.

What Does a Hedge Fund Research Desk Actually Look Like?#

What you lack is not information sources; you lack organization.

I break down the research desk into 9 roles (you can think of them as 9 positions):

Atlas (Research PM/Research Secretary): Receives inputBreaks down topic → Lists assets → Assigns questions
Macro: Rates / USD / Liquidity
Equity Sector: Sector rotation / Relative strength
Equity Stock: Company models / Catalysts / Risks
Crypto Market: Risk appetite / Narratives / Correlations
Derivatives: Funding / OI / Liquidation / Crowding
Risk: Portfolio exposure / Scenario stress testing / Risk control actions
Quant: Standardization / Trend momentum / Key level validation
CIO: Arbitrates evidenceTranslates into positioning and trading language → Outputs final report

How Did I Turn It Into an AI Workflow?#

In one sentence: One input → Seven parallel tracks → CIO consolidation → Single-page HTML report.

The workflow structure is:

Input → Atlas → (Parallel) Macro / Sector / Stock / Crypto / Derivatives & Onchain / Risk / Quant → CIO → Output

Tutorial: Get It Running in 5 Minutes (Truly Beginner Edition)#

You don't need to know how to code. Just copy.

Step 1: Create a Google Gemini Project#

Go to https://gemini.google.com/
Click on Gem in the left sidebar
Click New Gem

Now you can rely on Google Labs' experimental project to build any AI Mini Apps using prompts.

Step 2: Build the Workflow with Fixed Node Names and Connect Them#

It's recommended to copy the node names directly:

Input (User Input)
Atlas (Generate)
Macro (Generate)
Equity Sector (Generate)
Equity Stock (Generate)
Crypto Market (Generate)
Derivatives & Onchain (Generate)
Risk (Generate)
Quant (Generate, optional)
CIO (Output/Final Report)

Connection rules:

InputAtlas
AtlasAll analyst modules
All analystsCIO
CIOOutput

Open the Advanced Editor to edit or input the prompts all at once.

Step 3: The Prompts (Only 4 Major Blocks Needed)#

You just need to paste the following four sections into the corresponding nodes.

1/4 Fund Manager: Atlas (Distributes tasks based on user input)#

Function: Breaks down your input into a summary, asset list, which modules to activate, and key questions for the CIO to answer.

2/4: Analyst General Template (Paste at the top of each analyst node)#

Function: Unifies output format + enforces "having evidence."

Copy this section to Macro/Sector/Stock/Crypto/Derivatives/Risk (paste one copy for each):

3/4: Quantitative Analyst (Optional, paste into the Quant node)#

Enable this if you want a report that's "more like a fund's." You can skip it if you don't want complexity.

This part is best built with Cursor scripts + brokerage quantitative MCPs to achieve complex strategy-related calculations. We'll cover that later.

For now, you can create a free Finhub API and fill it into the prompt:

https://finnhub.io/

4/4: Chief Investment Officer (CIO) Final Investment Strategy Report#

Function: Arbitrates all modules into an "executable" research memo.

Final Step: Generate the mini app#

If you only use Gem's simple interface; you can input all the prompts at once into the dialog box to directly generate the mini app; then, in the app on the right, input any news, viewpoint, stock, or token code, and it will automatically output the report.

Click Download file: to download the HTML file and render the entire report directly in your browser.

This system is built on Google Labs' Opal, which is currently an experimental product and can occasionally be buggy or glitchy.

But for complete beginners, its greatest value can be summed up in one sentence: Without writing a single line of code, use prompts to first get the "research deliverable" working.

I'll cover the more hardcore parts later: Using Claude Code to generate scripts to complete data fetching, calculation, and verification; then packaging the entire process into a reusable skill, handing it over to OpenClaw's cron to run on schedule—you just need to set the topic, and it delivers the report to you on time.