SoftwareResearch/g2dm-research

G2DM Product Fitment Research Tool

Automates your G2 Digital Markets category research: pick a category, set the Special Considerations (include/exclude rules), paste your product rows, and get back qualify/reject decisions — each with a written reason and a clickable evidence trail — in the exact column order of your sheet.

The engine is Claude (`claude-opus-4-8`) with live web search + web fetch. It reads only the official vendor website to judge category fit, lifecycle (acquired/discontinued), Standalone-vs-Suite, and USD pricing. The verbatim <title> and <meta description> are extracted deterministically from the page.

Setup (one time)

1. 1.Get an Anthropic API key: https://console.anthropic.com → Settings → API Keys.
2. 2.In this folder, create your key file:

   cp .env.example .env
   # then edit .env and paste your key after ANTHROPIC_API_KEY=

Run

./run.sh

It creates a virtual environment, installs dependencies, and starts the app on a production server (gunicorn) at http://127.0.0.1:5057, opening your browser.

If .env contains APP_PASSWORD, the app asks for a login (username team, that password) even locally — that's the same gate used when hosting. Remove the APP_PASSWORD line from .env to run locally without a prompt.

To share it with your team for free, see [HOSTING.md](HOSTING.md).

How to use

Research tab

1. 1.Category (e.g. Payment Processing) and Market (default US).
2. 2.Special Considerations — your include/exclude rules. Click Review these rules to have the AI critique and suggest improvements before you run.
3. 3.Products — paste rows from your sheet, one product per line, tab- or comma-separated, in the order Vendor, Product, URL, Prior status:

   PayPal	PayPal	https://www.paypal.com	Pre-qualified
   Stripe	Stripe	https://stripe.com	Pre-qualified

1. 1.Run research. Watch live progress; click any row to expand its evidence.
2. 2.Export:
3. 3.Copy for Sheets — copies the fillable columns (tab-separated, no header) so you can paste straight into your sheet's columns.
4. 4.CSV / XLSX — full standalone file with headers and input columns.

Rows flagged ⚠️ are outliers (low confidence, or the signals-based and independent checks disagreed) — review those by hand.

Considerations Reviewer tab

Paste one or many categories to review their rules at once, using Category :: rules, separated by blank lines. You get an assessment, a list of issues, and a clean rewrite you can copy.

Files

• —app.py — Flask routes + streaming progress + XLSX export + password gate + health check
• —research_engine.py — per-product Claude research + JSON parsing + concurrency + retries
• —considerations.py — Special Considerations reviewer
• —html_extract.py — verbatim title/meta extraction
• —prompts.py — the research methodology + reviewer prompts
• —static/ — the web UI
• —gunicorn.conf.py — shared production server config (local + hosted)
• —run.sh — local production launcher · serve_public.sh — instant free public link
• —Dockerfile / Procfile — for free hosting (see HOSTING.md)

Notes

• —Runs locally; your key stays on your machine. Structured so it can be hosted for the team later.
• —Decisions use vendor sites only; third-party/review sites are excluded, and add-ons / plugins / integration-only products are rejected, per the methodology.