# AI use cases for Alpha Research OS (heranliu/AlphaResearchOS)

Repo: [heranliu/AlphaResearchOS](https://github.com/heranliu/AlphaResearchOS) | Use cases: 5 | Page: https://dedupedai.com/use-cases/heranliu-alpharesearchos/

AlphaResearchOS is a local, Python-based quant research workbench that takes a standard OHLCV CSV and runs an agent-assisted loop of hypotheses, predictive models, independent review, and holdout evaluation.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Run an AI-guided factor search on your own price CSV: https://dedupedai.com/use-cases/heranliu-alpharesearchos/run-first-ai-guided-factor-search/
- Audit the published sector ETF benchmark numbers: https://dedupedai.com/use-cases/heranliu-alpharesearchos/audit-sector-etf-benchmark-numbers/
- Stress-test a strategy against trading costs and turnover: https://dedupedai.com/use-cases/heranliu-alpharesearchos/stress-test-strategy-costs-and-turnover/
- Build a searchable factor library and export a research report: https://dedupedai.com/use-cases/heranliu-alpharesearchos/build-reusable-factor-library-report/
- Have an AI agent verify changes to Alpha Research OS: https://dedupedai.com/use-cases/heranliu-alpharesearchos/verify-changes-to-alpharesearchos-with-agent/

## How to run an AI-guided factor search on your own price CSV with Alpha Research OS

Repo: [heranliu/AlphaResearchOS](https://github.com/heranliu/AlphaResearchOS) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README

Import daily OHLCV data and let a model propose, train, and review candidate strategies.

**Who it is for:** A quant researcher or analyst with daily price data for a set of assets who wants to test strategy ideas without building the pipeline from scratch.

**The problem:** Turning a raw price file into tested trading hypotheses usually means writing feature code, model training, validation, and review logic by hand.

**The outcome:** A local research run that proposes feature expressions, fits predictive models, reviews each candidate, and shows development and holdout results in the workbench.

### Steps

1. Run `uv sync --locked` in the repository folder.
2. Run `uv run alphaos serve`, then open http://127.0.0.1:8765.
3. Click Import CSV / 导入 CSV, review coverage and validation results, and confirm the import.
4. Connect a model: choose your authenticated Codex CLI, a DeepSeek, Gemini, or Ollama preset, or a custom compatible endpoint, enter a model ID, save, and test the structured response.
5. Select the imported dataset, enter a research direction and budget (defaults are 6 candidates, 12 model requests, and 900 seconds), and follow the candidates as they are proposed, trained, reviewed, and evaluated.

### Prompt to paste into your agent

```text
In the AlphaResearchOS repo, start the workbench with `uv run alphaos serve`, import the CSV at <path-to-csv> with the standard columns, and start a research run with the default budget. Report the final candidate list with development scores and holdout results. Verify that every candidate has a recorded review decision before you report it.
```

### Caveat

The README describes the research loop and benchmark, but it does not promise that any strategy found by the search will hold up on new data.

Risk, credentials: Model API keys for DeepSeek, Gemini, or a custom endpoint are entered in the workbench, so store them carefully and do not paste them into shared logs or agent prompts.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heranliu-alpharesearchos/run-first-ai-guided-factor-search/

## How to audit the published sector ETF benchmark numbers with Alpha Research OS

Repo: [heranliu/AlphaResearchOS](https://github.com/heranliu/AlphaResearchOS) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Check the reported returns, Sharpe ratio, and drawdowns against the included result files.

**Who it is for:** A researcher or reviewer who needs to verify a published strategy benchmark before citing or reusing it.

**The problem:** Benchmark claims are only useful if the underlying data, methods, and derived metrics can be traced and checked.

**The outcome:** A reviewer who has traced each benchmark figure to its methods document and the stored result files, and can explain the setup behind each number.

### Steps

1. Clone the repository from https://github.com/heranliu/AlphaResearchOS and open its folder.
2. Read the methods, development results, and chart reproduction notes in docs/BENCHMARK.md.
3. Open benchmarks/sector-etf/results.json to find the immutable benchmark data.
4. Open benchmarks/sector-etf/analysis_metrics.json to compare the derived diagnostics with the figures in the README.
5. Run `uv sync --locked`, then `uv run alphaos serve`, and rerun the sector ETF research if you need to confirm the outputs from the source data. (inferred, check the README)

### Prompt to paste into your agent

```text
Using the AlphaResearchOS repo, read docs/BENCHMARK.md and benchmarks/sector-etf/results.json, then list each README benchmark figure next to its source value in the files. Report any figure you cannot find in those files, and verify that the holdout window dates match the README.
```

### Caveat

The benchmark covers nine US sector ETFs over one holdout window with 10 bps one-way costs, so the figures do not describe other markets or periods.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heranliu-alpharesearchos/audit-sector-etf-benchmark-numbers/

## How to stress-test a strategy against trading costs and turnover with Alpha Research OS

Repo: [heranliu/AlphaResearchOS](https://github.com/heranliu/AlphaResearchOS) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README

Check whether a strategy's edge survives higher transaction costs and heavier trading.

**Who it is for:** A portfolio analyst who has a candidate strategy and needs to know whether its returns depend on cheap trading assumptions.

**The problem:** A strategy that looks strong at low costs can lose its edge once realistic transaction costs and turnover are applied.

**The outcome:** A cost and turnover view of the strategy that shows how returns and turnover change as costs rise, next to equal-weight and momentum baselines.

### Steps

1. Run `uv sync --locked` and then `uv run alphaos serve` in the repository folder.
2. Open http://127.0.0.1:8765 and load a dataset that already has a candidate strategy from a completed research run.
3. Open the analysis view and check the equity, drawdown, and turnover charts for that strategy.
4. Change the one-way cost setting and compare the returns with the equal-weight and momentum baselines at each cost level.
5. Apply an executable turnover constraint if you want to test how a tighter trading limit changes the results.

### Prompt to paste into your agent

```text
In the AlphaResearchOS workbench, take the strategy named <strategy-name> and rerun its analysis at one-way cost levels of <cost-levels>. Return a table of total return, Sharpe ratio, and average turnover for each cost level and baseline. Verify that the baselines use the same cost setting as the strategy at each level.
```

### Caveat

The README describes cost and turnover analysis as features, but it does not state the exact cost model used beyond the benchmark's 10 bps and 40 bps one-way settings.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heranliu-alpharesearchos/stress-test-strategy-costs-and-turnover/

## How to build a searchable factor library and export a research report with Alpha Research OS

Repo: [heranliu/AlphaResearchOS](https://github.com/heranliu/AlphaResearchOS) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README

Keep useful features in one library and export a full report for review.

**Who it is for:** A research lead who runs several factor searches and needs to share results and reuse strong features across projects.

**The problem:** Factor ideas and their results get lost across runs, which makes it hard to reuse what worked or explain past decisions.

**The outcome:** A searchable factor library and an exported research report that together document the features, fitted-model settings, review decisions, and development results.

### Steps

1. Run `uv sync --locked` and then `uv run alphaos serve` in the repository folder.
2. Complete one or more research runs from the workbench on an imported dataset.
3. Open the factor library and search for features by name or by the research run that produced them.
4. Save the features you want to keep so they stay available for later runs.
5. Export the complete research report for the run and store it with the dataset and model settings used.

### Prompt to paste into your agent

```text
In the AlphaResearchOS workbench, search the factor library for features from <research-run-name> and export the full research report for that run. Return a summary of the kept features with their development scores. Verify that the exported report includes the fitted-model settings and review decisions.
```

### Caveat

The README says the library is searchable and exportable but does not describe its file format or how features move between separate installations.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heranliu-alpharesearchos/build-reusable-factor-library-report/

## How to have an AI agent verify changes to Alpha Research OS with Alpha Research OS

Repo: [heranliu/AlphaResearchOS](https://github.com/heranliu/AlphaResearchOS) | Difficulty: Advanced | Time: 45 minutes | Basis: From the README

Run the project's test, lint, script, and build checks before accepting a code change.

**Who it is for:** A developer contributing to Alpha Research OS who wants an AI coding agent to check a change before it is reviewed.

**The problem:** Changes to the research engine or the workbench can break tests, lint rules, or the packaged build if they are not checked.

**The outcome:** A change where the test suite, lint check, static script check, and package build have all been run and their results recorded.

### Steps

1. Run `uv sync --locked --extra dev` in the repository folder.
2. Run `uv run pytest -q` to run the test suite.
3. Run `uv run ruff check src/alpharesearchos scripts` to lint the code.
4. Run `node --check src/alpharesearchos/static/app.js` to check the workbench script.
5. Run `uv build` to confirm the package builds.

### Prompt to paste into your agent

```text
In the AlphaResearchOS repo, make the change described in <change-description>, then run the five development checks listed in the README and report each result. Verify that every check passes before you say the change is ready, and list any failure with its output.
```

### Caveat

These checks confirm the code runs and passes the listed tools, but the README does not say they verify the quality of the research results.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/heranliu-alpharesearchos/verify-changes-to-alpharesearchos-with-agent/