pickline-analyzer

Harvests dating-app openers from Reddit (or from a screenshot you feed it), scores each one with Claude on three dimensions, and ranks the top performers over a time window.

The question it answers: of the openers people actually post about, which ones correlate with a real response? Not a generator of cheesy lines — an analyzer of which ones land.

---

What it does

1. Harvest — pulls posts/comments from dating subreddits (Tinder, seduction, PickUpLines, hingeapp, Bumble, …) via Arctic Shift, then uses Claude Haiku to extract the actual opener, the reply (if any), and any outcome hint.
2. Score — Claude scores each line on creativity, response_quality, and hookup_probability:

   composite = 0.20·creativity + 0.30·response_quality + 0.50·hookup_probability
   

3. Rank — top-K over a window (today, 7d, all, …).
4. Screenshot mode — hand it a screenshot of a Tinder/Hinge conversation and Claude vision extracts the line + response + outcome, then scores it alongside the harvested corpus.

There's also a small web UI (web/bff + web/frontend) over the same store.

Quickstart

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env          # ANTHROPIC_API_KEY

python main.py harvest --window today
python main.py score
python main.py topk --window today --k 10
python main.py screenshot path/to/convo.png --platform Tinder
python main.py stats

Lineage — shared scraping harness

arctic_shift.py and store.py are intentionally the same Reddit-harvesting core used in my reddit-pushpull project. That one builds a relationship-dynamics corpus (ghosting, breadcrumbing, love-bombing); this one retargets the identical harness — Arctic Shift client + SQLite store — at dating openers and bolts a Claude scoring/ranking layer on top. Same plumbing, different question.

Data

The harvested corpus (picklines.db) and output/ are gitignored — only code ships. Keys come from .env (.env.example provided).